Ijciis July 2010 Vol. 1 No. 5

International Journal of
Computational Intelligence and
Information Security
ISSN: 1837-7823
July 2010
Vol. 1 No. 5
© IJCIIS Publication
International Journal of Computational Intelligence and Information Security, Vol. 1 No. 5, July 2010
IJCIIS Editor and Publisher

P Kulkarni
Publisher’s Address:
5 Belmar Crescent, Canadian
Victoria, Australia
Phone: +61 3 5330 3647
E-mail Address: ijciiseditor@gmail.com
Publishing Date: July 30, 2010
Members of IJCIIS Editorial Board
Prof. A Govardhan, Jawaharlal Nehru Technological University, India

Dr. Awadhesh Kumar Sharma, Madan Mohan Malviya Engineering College, India
Prof. Ayyaswamy Kathirvel, BS Abdur Rehman University, India
Prof. Deepankar Sharma, D. J. College of Engineering and Technology, India
Dr. D. R. Prince Williams, Sohar College of Applied Sciences, Oman
Prof. Durgesh Kumar Mishra, Acropolis Institute of Technology and Research, India
Dr. Imen Grida Ben Yahia, Telecom SudParis, France
Dr. Himanshu Aggarwal, Punjabi University, India
Dr. Jagdish Lal Raheja, Central Electronics Engineering Research Institute, India
Prof. Natarajan Meghanathan, Jackson State University, USA
Dr. Oluwaseyitanfunmi Osunade, University of Ibadan, Nigeria
Dr. Ousmane Thiare, Gaston Berger University, Senegal
Dr. K. D. Verma, S. V. College of Postgraduate Studies and Research, India
Prof. M. Thiyagarajan, Sastra University, India
Prof. Nirmalendu Bikas Sinha, College of Engineering and Management, Kolaghat, India
Dr. Rajesh Kumar, National University of Singapore, Singapore
Dr. Raman Maini, University College of Engineering, Punjabi University, India
Dr. Shahram Jamali, University of Mohaghegh Ardabili, Iran
Prof. Sriman Narayana Iyengar, VIT University, India
Dr. Sujisunadaram Sundaram, Anna University, India
Dr. Sukumar Senthilkumar, National Institute of Technology, India
Prof. V. Umakanta Sastry, Sreenidhi Institute of Science and Technology, India
Journal Website: https://sites.google.com/site/ijciisresearch/
2
Contents
1. A Secure Architecture Platform for Mobile Electronic Payment (pages 4-9)

2. Towards Improving E-Registration System using Email Technology and Knowledge
Management. (pages 10-16)
3. EFSM-Editor: An editor for fuzzy semantic data models (pages 17-27)
4. Credit Card Fraud Detection using Kernel in SVM (pages 28-33)
5. Attribute vector optimization for a non supervised classification by evolutionary
approach (pages 34-45)
6. Segmentation of Textured Images using Orientation as Feature Vector for Defect
Detection in Wood and Textile Industries (pages 46-53)
7. Fuzzy Time Series Forecasting accurately from Mean with Different Intervals (pages
54-62)
8. Recognition of Divergent Behaviour of a Plant Process Based on Direction of
Deviation of the Patterns Using Artificial Neural Networks (pages 63-70)
9. Application of Fuzzy Priority Scheduler for supporting QoS in MANET for DSR
Protocol (pages 71-76)
10. The Roll of Knowledge Shearing and Web-Based Applications on Learning Language
(pages 77-82 )
11. Behavioral model of 16 bit RISC Microcontroller in VHDL (pages 83-91)
12. Two Phases Vaccination for Network Worms (pages 92-98)
13. KMSVM model For Network Intrusion Detection (pages 99-105)
14. Improving Security and Reducing Computation Time using the ECDSA and PHAL
(pages 106-111)
15. Risk Assessment for E-commerce Security based on Fuzzy Iteration Model(pages 112-
118)
16. Analysis of Bridgeless PFC Boost Converter under CCM and DCM/CCM Boundary
(pages 119-125)
3
A Secure Architecture Platform for Mobile Electronic Payment
M. Udayapal Reddy1, R.K. Pateriya2, R.K. Baghel3

1
M. Tech (CSE) Scholar, Maulana Azad National Institute of Technology,
Bhopal-462051, India
2
Associate Professor, IT Department, Maulana Azad National Institute of Technology,
3
Associate Professor, Electronics Department, Maulana Azad National Institute of
Technology, Bhopal-462051, India
1
udayapalreddy@gmail.com, 2pateriyark@gmail.com, 3baghelrk@gmail.com
Abstract
The advent of the Internet has enabled new ways to conduct banking business, resulting in the creation of
new institutions, such as online banks, online brokers and wealth managers. Such institutions still account for a
tiny percentage of the industry. Over the last few years, the mobile and wireless market has been one of the
fastest growing markets in the world and it is still growing at a rapid pace. So in order to attract more customers
to such mobile applications, we need cent percent secure architecture platform for transaction into mobile
networks.In this paper a new Secure Model for Electronic Payment (SMEP) is proposed which fulfills most of
the requirements for secure trading on mobile in today’s scenario.
Keywords: SMEP, symmetric key, RSA, AES, wireless mesh network.
4
1. Introduction
WITH the hasty development and implementation of wireless mobile networks such as 3G UMTS[1] [2],
WiMAX[3] and Wi-Fi[4], have attracted more and more attention in our daily lives. Since the wireless network
are very stable, the users are now more interested in value added mobile applications over this wireless network.
Most of the manufacturers launch their product with mobile applications of electronic trading also called as
mobile commerce or m-commerce. Hence there is a need of good secure mobile trading model that must be
designed to attract more mobile users for doing business over a wireless network.
Therefore, how to combine the mobile applications with a secure mobile electornic trading model becomes
the prime factor as far as the design issues are concerned. This will greatly affect the success and performance of
any value added mobile applications for m-commerce. This is what the paper is concerned about.
2. Problem Definition
Mobile applications can be categorized into session-based applications and event-based applications. In
event-based applications, user’s payment is reflected by one-time events. Examples include sending a message,
querying traffic information, or purchasing a song. A session-based application consists of three phases: the
session-setup phase, the communication phase and the session release phase.
A customer is charged for a session-based application based on either time spent or data volume transferred,
e.g., VoIP-calling, video streaming, audio-streaming, or video-conferencing. There are a few payment models
proposed in the literature [5][6], which can be classified into two categories: the traditional payment model and
the micropayment model. The examples of traditional payment models include the credit card platforms
[7][8][9][10], and the electronic cash platforms[11][12][13]. The traditional payment models allow only one
payment in a payment transaction, which has been widely adopted for the event-based applications. Since a
session based application usually requires multiple payments during the execution of this application, with the
traditional payment model, it requires multiple payment transactions to complete a session-based application.
This is inefficient because heavy signaling and computational overheads are introduced into the network. On the
other hand, the micro payment models [14][15][16][17] allow multiple payments in a payment transaction,
which is considered more efficient than the traditional payment model. Thus, the micro payment models are
often adopted for most of mobile applications. To secure transactions, in the public-key cryptography [14][17]
(e.g., RSA[19]) is adopted. Unfortunately, the public-key cryptography requires heavy computation and long
execution time, which may not be a good solution in wireless mobile networks. Yang et al[16] applied the
symmetric-key cryptography such as Advanced Encryption Standard (AES) [20] that is more efficient than the
public-key cryptography in terms of computational cost and is more suitable for mobile devices. Unfortunately,
the symmetric-key cryptography requires more frequent key establishments and updates to prevent the shared
key from being compromised, and hence induces more communication cost due to key establishment and key
updates. Moreover, how to establish the shared key in wireless mobile networks for the symmetric-key
cryptography is very challenging. Compared with fixed networks, mobile networks have lower bandwidth,
longer transmission latency, and more unreliable connections, and mobile devices are restricted by limited
memory size and low CPU computational capability [21]. The installation of mobile applications on a mobile
network should be quick and of low cost.
3. Proposed System
To summarize, the following requirements should be addressed when designing a suitable trading
mechanism on a mobile network.
 First, customers expect a robust, secure, and fair trading mechanism which can be applied in different
mobile networks.
 Second, the trading mechanism should be light-weight (i.e., with low computational complexity and
low communication overhead) so that it can be easily run on mobile devices.
 Third, user anonymity should be achieved, that is, users’ purchasing behavior or preference should not
be traceable by others.
Finally, a trading mechanism should be of low implementation cost.
In this paper, we design an application-level secure payment model, named Mobile Electronic Payment
(SMEP), for wireless mobile networks, which attempts to meet these requirements. It is based on a more general
trading architecture model, which combines both public-key cryptography and symmetric-key cryptography to
5
overcome the disadvantages of both technologies. Specifically, we apply the emerging ID-Based Cryptography
in the SMEP to generate the public-private key pairs so that the certificate overheads among the network
operator (denoted as O), the user (denoted as U), and the mobile application developer or content provider
(denoted as P) commonly required in the traditional public-key cryptography can be eliminated. Then, from
these public-private key pairs, we generate three symmetric keys ku−o (held by O and U), ko−p (held by O and
P), and ku−p (held by U and P) to encrypt and decrypt the signaling messages exchanged among O, U, and P. An
important observation is that these three symmetric keys are established without actually exchanging them
among the concerned parties, a unique feature of ID-based cryptography. To prevent the symmetric keys [4] from
being compromised, in each payment transaction, the three public-private key pairs (kpub,o, kpri,o) held by O,
(kpub,p, kpri,p) held by P, and (kpub,u, kpri,u) held by U are used to generate the new symmetric keys. Our
design keeps the key freshness and thus provides more robust security protection. Moreover, SMEP supports
both event-based and session-based applications and is suitable for the resource-constrained mobile devices
because SMEP attempts to alleviate the computational cost and reduce the memory space requirement in mobile
devices. We expect that our SMEP provides a viable trading model for the future mobile applications.
4. Security requirements of Wireless Mesh Networks (WMN)

 Infrastructure security: This means the security of signaling and data traffic transmitted over the
infrastructure.
 Network access security: This indicates the communication security between a mesh client and a mesh
router.
 Application security: This refers to the security of mesh clients’ concrete data applications.
5. Secure Mobile Electronic Payment (SMEP) model Platform

In this section, we present the SMEP platform which follows the general trading model. When a new user U
or a mobile application/content provider P joins the SMEP, the Key Distribution procedure (to be elaborated
later) is executed to distribute U or P public-private key pairs denoted as (kpub,u,kpri,u) or (kpub,p, kpri,p),
respectively. Then, U can purchase a mobile application from P by running a payment transaction. In a payment
transaction, the signaling messages exchanged among O, U, and P are encrypted using three symmetric keys
ku−o (held by O and U), ko−p (held by O and P), and ku−p (held by U and P). The three symmetric keys are
updated (by utilizing the public-private key pairs) at the beginning of every payment transaction. A payment
transaction consists of three phases, the Withdrawal phase (where U obtains te tokens from O), the Payment
phase (where U uses the tokes to purchase a mobile application from P), and the Deposit phase (where P redeems
the obtained tokens from O).
In the following subsections, we first illustrate the key distribution procedure and then describe how a
payment transaction is executed in SMEP.
5.1. The key Distribution Procedure
The key distribution procedure generates public-private key pairs for O, U and P. The design of this
procedure utilizes the IBC [22] to eliminate the certificate overhead from binding one’s ID with its public key. In
the IBC, each user owns a key pair (kpub, kpri). The kpub is a public key, which is derived from a user’s identity
information (e.g., user’s telephone number or email address). The derivation of kpub can be done at the user’s
device or at the trusted authority (e.g., a network operator). The kpri is a private key, which is generated by the
trusted authority by taking the kpub into a function f and is passed to the user through a secure link.
 O first generates a public-params set (K, G1, G2, ˆe, kpub, o, H1, H2, H3) by the GENERATE-PARAMS
algorithm.
 O selects a random number S , and derives its private key kpri,o by computing kpri,o = S · kpub,o
 U sends O the UserAccountRequest message to apply for a user account.
 Upon receiving U’s request, O selects an IDu, and creates an account for U. Then O generates U’s
public key kpub,u and private key kpri,u then O sends kpub,u ,kpri,u and IDu to U through bearer network
link.
In the same way , P applies a third-party account by sending the Third Party Account Request message to O. By
6
generating public key and private key O sends k pub,p, k pri,p and IDp through the secure link between O and P.
5.2. Payment Transaction Procedure
A payment transaction consists of three phases[23], the Withdrawal phase (where U obtains tokens from
O), the Payment phase (where U uses the tokens to purchase a mobile application from P), and the Deposit phase
(where P redeems the obtained tokens from O).
1) Withdrawal Phase: In this phase, U obtains the electronic means (i.e., the tokens) from O. The message
flow for this phase includes following steps
 By browsing P’s website, U selects a mobile application, gets P’s ID, IDp, and obtains the Order
Information (OI) containing the ID and the data unit price of the mobile application. Then, U randomly
selects an integer Ru−o and generates the symmetric key ku−o. kpri,u)).Then U sends an
InitPaymentTrans message to O to initiate a payment transaction, information for O to generate the
tokens for U.
 Upon receipt of the InitPaymentTrans message, O will perform the authentication, if the authentication
is successful; O executes the GENERATE-TOKENS algorithm to generate N tokens. After executing
the algorithm, O sends U the TokenInfo meassage.
 Upon receipt of the TokenInfo message, U uses the token root to generate N tokens by executing the
GENERATE-TOKEN algorithm.
 O sends P a PurchaseInfo message to notify that U wants to purchase the mobile application[23].
 Upon receipt of the PurchaseInfo message,P stores the information required in the payment phase and
the deposit phase into its database.
2) Payment Phase: In the Payment phase, U uses the tokens to purchase a mobile application from P. This
phase may consist of one or more payments.
The message flow for the Payment phase includes following steps
 U initiates the first payment by sending an Initial Payment message to P, and the Initial Payment
message does not contain any information that may leak out U’s identity.
 Upon receipt of the Initial Payment message, P identifies the mobile application that U wants to
purchase and prepares data units to the mobile application. P encrypts the data units and responds with
Delivery message, to U.
 Upon receipt of the Delivery message, U decrypts the message and obtains the data units. Then, U sends
Payment message to P.
 Upon receipt of the Payment message, P decrypts the message. whether
Repeating the last two steps, U sends the succeeding tokens to P, and P delivers the succeeding data units to
U. This phase may be terminated if U stops paying the token or P stops delivering the mobile application.
3) Deposit Phase: Assume that P receives tokens after the Payment phase. In the Deposit phase, P redeems
the tokens from O.
This phase consists of the following two steps.
 P sends O the deposit message.
 Upon receipt of the deposit message, O deposits the credit for tokens into P’s account and takes the cost
for tokens from U’s account. The payment transaction is completed.
6. Conclusion
In this study we proposed a Secure Mobile Electronic Payment (SMEP) architecture model for the mobile
commerce over a wireless mobile networks. In this model, we take the advantage of the emerging ID-Based
Cryptography which eliminates the necessity of certificates commonly required by other public key
cryptography. Moreover, since ID-based cryptography can establish the shared key between two parties without
additional message exchanges, symmetric key cryptography can be still used effectively, leading to significant
computational cost. Our study shows that our SMEP model satisfies the requirements of secure trading such as
avoidance of overspending and double spending, fairness, user anonymity, and privacy and has low
7
computational cost. We expect that our SMEP model will provide a viable trading platform for the future mobile
applications and play an important role in the future emerging m-commerce industries.
Acknowledgement
The Success of this research work would have been uncertain without the help and guidance of a dedicated
group of people in our institute MANIT Bhopal. We would like to express our true and sincere
acknowledgements as the appreciation for their contributions, encouragement and support. The researchers also
wish to express gratitude and warmest appreciation to people, who, in any way have contributed and inspired the
researchers.
References
[1] K. Heikki, A. Ari, L. Lauri, N. Siamak, and N. Valtteri, 2002 ,“UMTS Networks- Architecture, Mobility
& Services”, John Wiley & Sons, Inc.
[2] Y.-B. Lin and I. Chlamtac, 2001, “Wireless and Mobile Network Architectures”, John Wiley & Sons, Inc.
[3] IEEE Std. 802.16-2004, 2004, IEEE Standard for Local and Metropolitan Area Networks-Part 16: “Air
Interface for Fixed Broadband Wireless Access Systems, technical report”.
[4] IEEE Std 802.11-1997, Information Technology-Telecommunications and Information Exchange between
Systems-Local and Metropolitan Area Networks-Specific Requirements-Part 11: “Wireless LAN Medium
Access Control (MAC) and Physical Layer (PHY) Specifications”, technical report IEEE.
[5] N. Asokan, P. A. Janson, M. Steiner, and M. Waidner, Sept. 1997, “The state of the art in electronic
payment systems,” IEEE Computer, vol. 30, no. 9, pp.28–35.
[6] Z.-Y. Lee, H.-C. Yu, and P.-J. Ku, 2001 “An analysis and comparison of different types of electronic
payment systems,” in Proc. Portland International Conference on Management of Engineering and
Technology, vol. 2, pp. 38–45.
[7] M. Bellare, J. Garay, R. Hauser, A. Herzberg, H. Krawczyk,M. Steiner,G. Tsudik, E. Herreweghen, and M.
Waidner, Apr. 2000 “Design, implementation and deployment of a secure account-based electronic
payment system,” IEEE J. Select. Areas Commun., vol. 18, pp. 611–627.
[8] M. M. Anderson, Oct. 1998, Financial Service Markup Language (FSML) version 1.17.1, “technical report
Financial Services Technology Consortium”.
[9] Mastercard and Visa, SET Secure Electronic Transactions Protocol, version 1.0 edition, May 1997, Book
One: “Business Specifications”, Book Two: “Technical Specification”, Book Three: “Formal Protocol
Definition”.
[10] S. Low and N. Maxemchuk, 1994, “Anonymous credit cards,” in Proc. 2nd ACM Conference on Computer
and Communications Security, pp. 108–117.
[11] J. P. Boly, A. Bosselaers, R. Cramer, R. Michelsen, S. F. jolsnes, F. Muller, T. P. Pedersen, B. Pfitzmann,
P. Rooij, B. de, Schoenmakers, M. Schunter, L. Vallee, and M. Waidner, 1994 “The ESPRIT Project
CAFE-high security digital payment systems”, in Proc. ESORICS, pp. 217–230.
[12] F. Medvinsky and B. Neuman, 1993, “NetCash: a design for practical electronic currency on the Internet,”
in Proc. First ACM Conference on Computer and Communications Security, pp. 102–106.
[13] J. Camenisch, U. Maurer, and M. Stadler, 1996, “Digital payment systems with passive anonymity-
revoking trustees,” in Proc. ESORICS, pp. 33–43.
[14] S.-M. Yen, Nov. 2001, “PayFair: a prepaid Internet micropayment scheme ensuring customer fairness,” in
Proc. IEE Computers and Digital Techniques, vol. 148,no. 6, pp. 207–213.
[15] A. Herzberg and H. Yochai, Apr. 1997, “Mini-pay: charging per click on the Web,” in Proc. Sixth
International World Wide Web Conference, Santa Clara, CA, pp. 301–307.
[16] Z. Yang, W. Lang, and Y. Tan, 2004, “A new fair micropayment system based on hash chain,” in Proc.
IEEE International Conference on e-Technology, e-Commerce and e-Service, pp. 139–145.
[17] R. L. Rivest and A. Shamir, 2003 “PayWord and MicroMint: two simple micropayment schemes,” IEEE
Trans. Veh. Technol., vol. 52, no. 1, pp. 132–141.
[18] J. Jonsson and B. Kaliski, 2003, “Public-key cryptography standards (PKCS) #1: RSA cryptography
specifications version 2.1, technical report RFC 3447.
[19] J. Daemen, S. Borg, and V. Rijmen, 2002, The Design of Rijndael: AES – “The Advanced Encryption
Standard”.
[20] Y. C. Lai, P. Lin, and Y.-T. Huang, Jan. 2006 “Design and implementation of a wireless Internet remote
access platform,” J. Wireless Commun. And Mobile Computing, vol. 6, no. 4, pp. 413–429.
[21] D. Boneh and M. Franklin, 2001 “Identity-based encryption from Weil pairing”, in Proc. Crypto 2001, pp.
213–229.
8
[22] Phone Lin, Senior Member, IEEE, Hung-Yueh Chen, Yuguang Fang, July 2008, Fellow, IEEE, Jeu-Yih
Jeng, and Fang-Sun Lu,” A Secure Mobile Electronic Payment Architecture Platform for Wireless Mobile
Networks”, IEEE Transactions On Wireless Communications, Vol. 7, No. 7.
9
Towards Improving E-Registration System Using Email Technology

and Knowledge Management
Hamdan.O.Alanazi 1, Rami Alnaqeib 1, M.A.Zaidan 1, A.A.Zaidan 2, B.B.Zaidan 2,

Omar Noori 3
1
Faculty of Computer Science and Information Technology / University of Malaya / Kuala
Lumpur/Malaysia
2
Department of Electrical and Computer Engineering/ Faculty of Engineering, / Multimedia
University / 63100 Cyberjaya / Selangor Darul Ehsan / Malaysia
3
Faculty of Information Technology, University Tenaga National, 43009, Km7 Jalan Kajang-Pochong,
Malaysia
Email Corresponding Author: aws.alaa@gmail.com
Abstract
While education sector grows rabidly the importance of having unique system, that satisfies both students
and registration department, became crucial. To understand the appropriate technologies that play significant role
in forming an innovative e-registration system, this paper aims to provide an understanding of the factors that
contribute in providing the students with an easy-to-use and to entry point to complete all departments as the
fuclty of computer science and information Technology in University of Malaya (FSKTM) as online
application. In this paper a new study is continuing to have investigations with features of e-registration systems,
and it opens the doors to ideas that would stimulate both students and university to reap the real benefits of e-
registration using email technology and Knowledge Management. To be more simply.
Keyword: - E-Registration, Knowledge Management, Elecronic Email Technology.
10
1. Introduction
Recently in 21st century, the rapid development of information technology and the rapid of increase in
information exchange have brought new drives and innovation ideas to the whole society, Web-Based
Application; present an introduction to the concepts, practices, theories, languages, and tools used to develop
applications for the internet. It explores the important of Web applications while also introducing the
programming and technology essential to put these ideas into practice [7], however there are several applications
poor to the pre-study, which present the backbone of any web-based application, therefore this study has focused
on the e-registration with System email technology and Knowledge Management.
2. Related Work
With the rapid growth of the Internet, organizations are constantly finding new ways to leverage this
technology as a means of achieving their objectives [10]. These objectives vary and can include enhancing client
relationships, promoting customer loyalty, streamlining internal communication and many other endeavours
[1],[2],[3],[4],[5],[6]. Although the goals may differ, the Internet has become one of the most essential
components of an organization's communication strategy. The Electronic mail, sometimes called email, is a
computer based method of sending messages from one computer user to another. These messages usually consist
of individual pieces of text which you can send to another computer user even if the other user is not logged in
(i.e. using the computer) at the time you send your message. The message can then be read at a later time. This
procedure is analogous to sending and receiving a letter. Originally, email messages were restricted to simple
text, but now many systems can handle more complicated formats, such as graphics and word processed
documents [8]. When mail is received on a computer system, it is usually stored in an electronic mailbox for the
recipient to read later. Electronic mailboxes are usually special files on a computer which can be accessed using
various commands [9]. Each user normally has their individual mailbox. It is straight forward to send electronic
mail between users of different computer systems which are connected to major networks. Most major academic
and research institutions and companies throughout the world can now be reached by electronic mail. In
addition, a growing number of individuals can be contacted in this way. In the UK, most academic and research
institutions are linked by a network called JANET (or Super JANET). This is effectively part of the Internet, so
email can be exchanged with most national and international networks. The World Wide Web has long been
considered the focal point of a successful Internet strategy. Organizations are, however, increasingly recognizing
the necessity of electronic communication to efficiently reach their goals. The ability to interact and
communicate with different audiences - such as customers, employees, partners and suppliers - via email,
quickly and easily, is now just as vital for an organization's success. This is where email lists come in. Email lists
are a very successful tool for achieving a variety of objectives. Email lists enable fast, global and economical
group communication. By using email lists, organizations can quickly deliver large numbers of email messages
for a fraction of the cost that other communication channels require [9]. This introductory white paper discusses
email lists and how to effectively use them to further your organization's goals. It explains what email lists are,
how different types of email lists are set up, and most importantly, how to effectively use them and make them
work for your organization. The operating principle behind email is relatively simple, which has quickly made it
the most popular service used on the Internet. As with a traditional postal service, for your message to reach your
recipient, all you need to know is their address. Its two main advantages over "paper mail" are the speed at which
the email is sent (practically instantaneous) and the lower cost (included with the cost of an Internet connection).
What's more, email can be used to instantaneously send a message to several people at once. Knowledge
management and E-registration online in FSKTM are examples for supported activities. Knowledge management
applications are aimed to provide organizations with tools to manage their business knowledge, while the focus
of this website has always been on managing the delivery of academic knowledge. Efforts to integrate both areas
of research are lacking. This website presents a knowledge management approach for knowledge management
applications. It highlights the joint characteristics of the two concepts and proposes a KM view of E-registration
online. The aim is to streamline the transfer of comments and questions and content among FSKTM students of a
typical administrators environment. The participants in this website depends on its ability to acquire and convert
the abilities and intellect of sharing information together and the skills with which administrator in the
faculty[11]. They have emerged as a result of the rapid penetration of technology in the delivery of opinions to
gather. “The real promise of comments and participants in opinions in its potential to enhance the learning
experience for learners” [12]. However, although the topic of courses in the website has been subjected to
significant research in many countries over the last two decades.
11
3. E-Registration and Knowledge Management Parts

This website (E-Registration online ) systems have been developed parts of knowledge management and
illustrated some principles of knowledge sharing that as a will appeared in the website E-Registration online . in
this website apply knowledge sharing between students and administrates when usage email technologe . The
website the students can be ability to write various knowledge and questions and opinions and also the student
can receive and answer the students in the website this consider parts of knowledge management. when the
students have an experience in his mind this considered as tacit knowledge.
We capture this tacit knowledge when he express or convert this tacit knowledge into explicit knowledge in
the form of text when he adds a topic or a comment. this process is called externalization the internalization
process take place when another user read this topic or comment what is now in the form of explicit knowledge ,
when he memorize this topic or comment , actually he is converting the explicit knowledge into tacit knowledge
and the domains and promote the activities of learners [14]. This technical support to learning is briefly
described as knowledge management part in this website. In other words, it refers to knowledge management
that will illustrate in this website where participants between students and administrates plays a major role in the
delivery of the ‘Opinions and questions pertaining to this website content’. Initially, the E-Registration online
have content is delivered via a range of technologies such as internet. However, the advances in communication
technologies and the growth in the use of the Internet have massively reshaped site capabilities because the
Internet enables greater interaction between students, administrate and the comments and questions contents. The
Internet also enables learning materials to be accessed from anywhere through the web or Intranet. In addition to
the usual face-to-face dialogue, administrates and students can also communicate with each other using e-mail,
chat or discussion forums. Such highly interconnected environment helps make students active seekers rather
than passive recipients of knowledge. This super connectivity also “enables education to occur in places where it
normally does not, extends resources (information) where there are few, expands daily participants and opens the
knowledge sharing information place as it connects people [11],[14].
4. Knowledge Management
Defining knowledge management (KM) has become rather like seeking a definition for ‘international
terrorism’. Everybody can talk about it but no one can define it precisely. In other words, there is no universal
definition of what knowledge management really is. This could largely be attributed to the differences in old
philosophers’ arguments about what constitutes “knowledge” in the first place. Various KM definitions can be
found in the literature. This could be justified by the multidisciplinary backgrounds of KM researchers. The
disciplines of KM researchers include management sciences, software engineering, artificial intelligence,
sociology etc. However, in a broad context, knowledge management can be defined as the process of capturing,
codifying, classifying, storing and sharing knowledge. Basically, the general scope of KM is to facilitate
organizational learning where knowledge workers are supported to capitalize and share their know-how. The aim
is to capture, store and improve the transfer of know-how within an organization. In order to fulfil these
promises, different KM strategies have been introduced. The main KM strategies to be employed by early
adopters of the princple [15] are:
 The process-centred approach mainly understands KM as a social communication process. In this
approach, knowledge is closely tied to the person who developed it and is shared mainly through
person-to-person contacts. The main purpose of information technology in this approach is to help
people communicate knowledge, not to store it. This approach is also referred to as the
“personalization” approach.
 The product-centred approach focuses on knowledge documents, their creation, storage and reuse in
computer-based corporate memories. This approach is also referred to as the “content-centred” or
“codification” approach.
In the following section, we highlight the similarities and shared goals of the knowledge management and
e-Registration online in the faculty of computer sciences and information technology and processes and how to
enrich the cause of sharing knowledge or information between students and administrators through the
incorporation of KM practices.
5. Conjunctive Relations and the Synergic Solution

During the last decade, both KM and Forum in E-Registration online have experienced many innovations
but are still largely unconnected [16]. Traditionally KM has been associated with the corporate students and
administrators; whereas Forum in E-Registration online has been traditionally applied more widely including the
participants or sharing information between them [17]. Though both concepts are intended to be very different
12
approaches to different topics, we believe that these approaches have a common ground in terms of their goals
and techniques. Basically, the general scope of KM is to eventually lead to establishing what could be regarded
as the “organisational brain” or technically called the organisational memory system (OMS). On the other hand,
the scope of this forum website is to help student’s brains in constructing and gaining prescribed ‘educational
knowledge’. If we look at the comments and information in the forum it has been concept in an abstract way, we
can easily identify certain KM activities which are embedded in the Website registration online Firstly, since its
inception, forum website has been regarded as the process of maintaining That contains a database that is stored
for information and comments that occur between students and the administrators.
This goal or the purpose of this participation to reach the right solutions to be agreed between the people
who use this site” Knowledge management is more than creating a knowledge archive and invoking retrieving
mechanisms” [18]. It is also aimed at providing logistics for knowledge transfer (i.e. personalization strategy).
Similarly, in this forum is also expected to provide communication between students, administrators and
themselves. For this reason, KM techniques can be used to provide logistic means for the transfer of sharing
knowledge. This includes facilitating learners’ monitoring and intercommunications management between
learners and instructors. Notice that this view is also reflected in defining KM as: “getting knowledge from those
who have it to those who need it” [19]. This view also matches the personalization strategy of KM.
Therefore, the effectiveness of incorporating the knowledge management approach in Forum website is
providing customized sharing knowledge cycles that suit the various[20],[21].
Figure 1: KM Framework of FSKTM E-Registration Online Solution
6. Implementation Scenarios
In order to realize the KM-based system approach for FSKTM, a set of learning cycles were defined. The aim is to
set up a workflow management within the forum framework. Figure 1 shows the framework that describes the defined
relationship between students and administrators in the faculty and feedback cycles. Firstly, the students write some
opinion about something related to website. The admin also write Questions or comments this is called knowledge
sharing.
7. Initials Assessment
Technology quality, reliability, usability and accessibility are important determinants within any forum environment.
The FSKTM management and the technical staff pays, some Impacts that the forum in FSKTM has:
 The use of the forum has changed the personality of students to become more confident.
 The forum materials help students to understand or to reach for solution about this website the topics easily.
 The use of forum expands general knowledge
 The use of forum helps to enhance self-learning skills.
 Knowledge gained through forum is more truthful and consistent compared to traditional learning.
 The use of Multimedia helps to have better view and understand topics easily the use of computers helps to
improve research skills.
13
8. Methodology
i. System Email Technology concept
All email systems have the ability to send, receive and discard mail. Most systems have facilities for storing mail which is
to be kept rather than discarded. It is important to discard mail which does not need to be kept, as it uses storage space on
disks. Mailboxes can soon accumulate a large number of mail messages making it difficult to read and process new mail,
in addition to wasting disk space. There is almost always a connection between the email system and the computer's
standard file system which allows mail to be read from files or written to files. This enables greater flexibility in how the
mail system is used. For example, a mail message may be prepared in a normal file using a familiar text editor and then
sent by the email system. Sections of other files may be included in the mail message as well. Most systems have a reply
facility, although some of these do not always work as expected. Care should be taken when using this facility in
electronic mail, as replies do not always go back to the sender. He recipient of mail may not always be an individual, but
could be a service such as Helpdesk, Postmaster, a mailing list or an automatic processing service. Mailing lists are
supported by many systems. These allow mail which is sent to the name of the list to be sent automatically to all addresses
in that list. In this way mail can be sent to one or more groups of users who share a common interest, e.g. members of a
user group or research team, by sending a single message. A number of information services are also available through
electronic mail whereby the mail is processed and answered by an automatic process on the remote system
[1],[2],[3],[4],[5],[6].
ii. System E-mail Structure
Every e-mail message has two parts: a header and a body. The message’s header specifies control information such as its
source and destination; the time it was mailed, and the person that created the message. The message’s body contains the
content of the message which might be plain text, or MIME (Multipurpose Internet Mail Extensions). MIME allows a
sender’s mail server to encode 8-bit binary files using a mail system that supports only plain text. A MIME-enabled mail
server encodes binary data as text and then the recipient’s mail server decodes the text back to binary file
[1],[2],[3],[4],[5],[6].
iii. Concept of E-mail
SMTP (Simple Mail Transfer Protocol) is a standard protocol which is used to transport e-mail. Actually, there are some
mail agents handle all the networking required to move mails around a network and they run as a background process.
However, when you are using a PPP connection to the internet, your internet mail computer uses SMTP to deliver mail
and POP3 (Post Office Protocol version 3) to receive mail. Figure 2: Concept of E-mail connection [9].
Figure 2: Concept of E-mail connection
iv. System E-mail Programs
There are some common popular e-mail programs include Eudora Light, Netscape Mail and Outlook Express. All these
programs have roughly the same features, and allow you to create, send, receive and organize your messages. In
particular, Outlook Express will be chosen as a demonstration in the following discussion.
14
v. System E-mail Addresses
You can send a message to anyone around the world if you know the person’s e-mail address. Furthermore, to get an e-
mail, you must have an e-mail address for the message to get in. An e-mail address defines the location of an individual’s
mailbox on the internet. The e-mail address has two parts: the first part is the domain name of the host computer on which
you check your mail, and the second part is the user name of the mail account, and the two parts are separated with an
ampersand (@), and an e-mail address cannot contain spaces. Domain name: the location of the person’s account on the
Internet. Periods (.) separate the various parts of the domain name. Username the name of the person’s account, this may
be a real name or a nickname (i.e. kyc@cuhk.edu.hk).
9. Conclusion
In this paper has been presented importance of email technology and its relationship to the Knowledge Sharing where we
explain the vocabulary of the email address of Technology aspects: System Email Technology concept, System E-mail
Structure, Concept of E-mail, System E-mail Programs, System E-mail Addresses and normalize the concept of
information has been shared and how it relates with the E-Registration System to provide an electronic system developed
and has been easy to use for the student.
References
[1] Ibrahim A.S.Muhamadi, A.A Zaidan, S.Raviraja, M.A Zaidan, B.B Zaidan, Chengetai Mapundu, “ Auto
Notification Service for the Student Record Retrieval System Using Short Message Service (SMS) ”,
International Journal of Computer Science and Network Security (IJCSNS) , Vol.9, No.8 , ISSN : 1738-
7906, pp. 200-208, 30 Augest (2009), Seoul, Korea.
[2] Ibrahim A.S.Muhamadi, A.A Zaidan, S.Raviraja, M.A Zaidan, B.B Zaidan, Chengetai Mapundu, “ Auto
Notification Service for the Student Record Retrieval System Using Electronic Mail Message (EMM) ”,
International Journal of Computer Science and Network Security (IJCSNS) , Vol.9, No.8 , pp. 295-303,
ISSN : 1738-7906, 30 Augest (2009), Seoul, Korea.
[3] Ibrahim A.S.Muhamadi, S.Raviraja, A.A Zaidan, M.A Zaidan, B.B Zaidan, Chengetai Mapundu, “
Enhanced Admin Role of the Student Record Retrieval System using Knowledge-Sharing and Rationale
Method“ ”, International Journal of Computer Science and Network Security (IJCSNS) , Vol.9, No.8 , pp.
269-276, ISSN : 1738-7906, 30 Augest (2009),Seoul, Korea.
[4] Ibrahim A.S.Muhamadi, M.A Zaidan, A.A Zaidan, B.B Zaidan, “Student Record Retrieval System Using
knowledge Sharing”, International Journal of Computer Science and Network Security (IJCSNS) , Vol.9,
No.6 , ISSN : 1738-7906, pp. 97-106, 30 June (2009), Seoul, Korea.
[5] Ibrahim A.S.Muhamadi, S.Raviraja, B.B Zaidan, A.A Zaidan, M.A Zaidan, Chengetai Mapundu, “ New
Quantitative Study for the Student Record Retrieval System ”, International Journal of Computer Science
and Network Security (IJCSNS) , Vol.9, No.8 , ISSN : 1738-7906, pp. 304-311, 30 Augest (2009), Seoul,
Korea.
[6] Ibrahim A.S.Muhamadi, M.A Zaidan, A.A Zaidan, B.B Zaidan, “Student Record Retrieval System”,
Proceeding of World Academy of Science Engineering and Technology (WASET), ), indexed by Thomson
Reuters (ISI), SCIRUS, Google Scholar, Engineering Index (Compendex), EBSCO, GALE, DOAJ,
INTUTE, Scientific Commons and Electronic Journals Library), Vol.56, ISSN:2070-3724, P.P 503-510, 26-
28 August (2009), Singapore
[7] Grove, R. F. (2009). Web-Based Application Development, Jones & Bartlett Pub.
[8] Murray, D. E. (1988). "The context of oral and written language: A framework for mode and medium
switching." Language in society 17(3): 351-373.
[9] Picard, D. F., T. L. Root, et al. (2001). System for accessing multimedia mailboxes and messages over the
internet and via telephone, Google Patents.
[10] Ranchhod, A. (2004). "The changing nature of cyber-marketing strategies." Business Process Management
Journal 10(3): 262-276.
[11] Wijetunge, P. (2002). "Adoption of Knowledge Management by the Sri Lankan University librarians in the
light of the National Policy on University Education." International Journal of Educational Development
22(1): 85-94.
[12] Li, W. (2001). Constructivist learning systems: A new paradigm. IEEE International Conference on
Advanced Learning Technologies, Madison, USA, IEEE.
[13] Mohamed, A. H., R. A. S. Abuzaid, et al. (2008). "Opportunities AND Challenges OF THE Knowledge
Management Approach TO E-LEARNING: AC ASE Study IN Al-Bayan Model School FOR Girls,
Kingdom OF Saudi Arabia." EJISDC 35(4): 1-11.
[14] Watanabe, T. (2003). Computer supported educational environment. IEEE Telecommunications, IEEE.
[15] Mentzas, G., D. Apostolou, et al. (2001). "Knowledge networking: a holistic solution for leveraging
corporate knowledge." Journal of Knowledge Management 5(1): 94-106.
[16] Althoff, K. D. and D. Pfahl (2003). "Making Software Engineering Competence Development Sustained
through Systematic Experience Management." Managing Software Engineering Knowledge, A. Aybe, R.
Jeffrey, C. Wohlin, M. Handzic (Eds.), Springer, Heidelberg/New York.
15
[17] Zuga, B., I. Slaidins, et al. (2006). "M-learning and Mobile Knowledge Management: Similarities and
Differences." International Journal of Computing & Information Sciences 4(2): 58?2.
[18] McBriar, I., C. Smith, et al. (2003). "Risk, gap and strength: key concepts in knowledge management."
knowledge-based Systems 16(1): 29-36.
[19] Fahad H. Alshammari, Rami Alnaqeib, M. A. Zaidan, Ali K. Hmood, B. B. Zaidan and A. A. Zaidan , “
New Quantitative Study for Dissertations Repository System ” , Journal of Computing (JOC), Vol.2, Issue
6, ISSN: 2151-9617, P.P 85-93, 30 June 2010, New York, USA.
[20] Fadhel.S.AlAbdullah, Fahad H. Alshammari, Rami Alnaqeib, Hamid A. Jalab, A. A. Zaidan and B. B.
Zaidan, ” Analytical Study on Internet Banking System ”, Journal of Computing (JOC), Vol.2, Issue 6,
ISSN: 2151-9617, P.P 140-146, 30 June 2010, New York, USA.
[21] Rami Alnaqeib, Fahad H. Alshammari, M. A. Zaidan, A. A. Zaidan, B. B. Zaidan and Zubaidah M. Hazza ”
An Overview: Extensible Markup Language Technology”, Journal of Computing (JOC), Vol.2, Issue 6,
ISSN: 2151-9617, P.P 177-181,30 June 2010, New York, USA.
16
EFSM-Editor: An editor for fuzzy semantic data models
Noura Labidi, Lassaad Ellouze and Rafik Bouaziz
MIRACL, Multimedia InfoRmation systems and Advanced Computing Laboratory

FSEGS, Route de l’Aérodrome, BP 1088, 3018 Sfax, Tunisia
noura.labidi@yahoo.fr; las_ellouze@yahoo.fr; raf.bouaziz@fsegs.rnu.tn
Abstract
This paper shows how we can model and implement a framework to build fuzzy semantic models based on
EFSM (Enhanced Fuzzy Semantic Model). More precisely, it presents a process, in conformity with the MDA
approach, for developping a graphical editor which must allow the representation and the management of EFSM-
based models. EFSM-Editor, the result of our work, keeps trace of these models under XMI format. Moreover, it
provides a code generator which transforms the graphical representation into the object code of EFSM.
Keywords: Fuzzy data, Fuzzy Semantic Model, Complex object, MDA approach, Graphical editor.
17
1. Introduction
The fast technological advances have led to new applications requiring the mastery of complex objects and
semantic relationships. Among these applications, there are those which manipulate vague, imprecise and
uncertain information. It is that the ignorance of the fuzzy nature of the real-world undoubtedly introduces
deformations of human perception. These applications are intensive data and require a suitable logic and
intelligent means of representation and processing.
This is why several works [1] have been devoted to the extension of classical models to incorporate fuzzy
data in databases. In this context, we find the extension of USM (Unified Semantic Model), which allows the
modeling of the object oriented and semantic databases [2]. This extension, entitled FSM (Fuzzy Semantic
Model) [3], authorizes an entity to be, partially, a member of its class according to a given Degree Of
Membership (d.o.m), taking a value between 0 and 1. This degree reflects the level to which the entity verifies
the properties of this class. Recently, FSM has been enriched by new concepts [4, 5] to allow the
parameterization of decision rules used to calculate the d.o.m of the entities, on the one hand, and the
consideration of certain semantic characteristics for better respecting the constraints of the real-world, on the
other hand.
Nevertheless, this extension of FSM, entiteled EFSM (Enhanced FSM), suffers, like the large majority of
the fuzzy models, from the absence of a tool CASE helping modeling of fuzzy semantic data. This paper focuses
therefore on defining and implementing an editor for the representation of fuzzy semantic models based on
EFSM. This editor is built in accordance with the MDA (Model Driven Architecture) approach.
The rest of the paper is structured as follows. We present in the next section the related work. Section 3
gives a brief presentation of EFSM. Section 4 describes our solution to implement the EFSM editor. Finally, the
last section presents our conclusions and perspectives.
2. Related work
For a few years, several research studies have been devoted to the integration of fuzzy logic in modeling
Data Bases (DB), as well relational [6, 7, 8] that object-oriented [3, 4, 5, 9, 10, 11] DB.
For the Fuzzy Relational DB (FRDB), two main approaches are to be distinguished. One is based on the
fuzzy relation and the relation of similarity [8], and the other is based on the possibility distribution [7]. Some
works introduce the fuzziness only on the level of the attribute [6], while others also incorporate it in the level of
the tuple [7, 8]. However, even with the fuzzy extension, these models are unable to model complex objects.
Concerning the Fuzzy Object-Oriented DB (FOODB), fuzziness has also been applied in object instances
and class hierarchies in order to handle uncertain data of complex objects. These extensions have resulted in
different approaches for conceptual modeling of fuzzy data [3, 4, 5, 12, 13, 14, 15], with appropriate graphical
representations. Most models are extensions of earlier ones, such as the ER model, used in the fuzzy extension
defined in [16], EER model, used in the extensions defined in [12, 13, 17], or IFO object model, used in the
extensions defined in [14, 15, 18]. Other extensions [9, 10] are based on the standard object data model of
ODMG (Object Data Management Group). These approaches are used to define degrees of membership to some
elements of the model and suggest membership functions of these degrees. But only the EFSM model [3, 4, 5]
provides graphical representations for decisional elements based in appropriate rules to determine these degrees.
EFSM is also distinguished by new concepts that allow modeling reality more accurately.
3. Presentation of EFSM
In addition to taking into account the fuzziness at the attribute level, as done in most other proposals, FSM
incorporates this concept in class level by allowing an entity to verify partially the properties of a class. This
refers to the degree of membership of an entity to a class. A fuzzy class K is defined in FSM as a collection of
fuzzy entities (or objects):
K = {(e,  K (e) ): e  E ^  K (e) > 0}. E is the set of all entities of the interest domain,  K is a characteristic or
membership function associated to K and  K (e ) represents the degree of membership (d.o.m) of a fuzzy entity e
in a fuzzy class K. The membership function  K maps the elements of E to the range [0, 1], where 0 implies no-
membership and 1 implies full membership. A value between 0 and 1 indicates the extent to which a fuzzy entity
e can be considered as an element of a fuzzy class K.
In FSM, we denote by XK = {p1, p2, …, pn} (with n  1) the set of properties for a given fuzzy class K. The
degree with which each of the extent properties determines a fuzzy class K is not the same. Indeed, there are
some properties that are more discriminative than others. To ensure this, this model associates to each extent
property pi a non-negative weight wi reflecting its importance in deciding whether or not an entity e is a member
of a given fuzzy class K , with: in1 wi  0 .
18
The d.o.m of a fuzzy entity e in a fuzzy class K is then calculated as follows:

 in  i ( i )   i
p k (1)
 K (e ) 
 in i
With  p i ( i ) is the partial membership function relating to the property pi.

k
It is a generic formula, used as a base for the definition of the membership rules of various elements of FSM
(association, aggregation, subclass/superclass relationship, composite class) [3, 19].
Although FSM is to our knowledge the most advanced model in the definition of decision rules, it has
insufficiencies. Thus, it has been enriched [4, 5] to take into account the impact of a simultaneous presence of
certain decision rules, of the inhibition of the effect of a rule by the presence of another, of multiple and / or
optional presence of a decision rule, and an alternative choice between a set of decision rules. Such an impact is
on the d.o.m of an entity to a class. Thus, new concepts were added to FSM to model with more fidelity the
needs of the real world.
 Decisional grouping concept: This new concept allows according a common weight to several decision rules
which must be applied globally on the same class to act as a composed decision rule with a common weight.
The decisional grouping concept is graphically represented by a node related by an arc to the considered class
and connected by line segments to the properties that it groups.
 Concept of decisional inhibition: This concept consists in inhibiting the effect of a decision rule or a group
of decision rules in the computing of the membership degree of an entity e in a class. It requires the definition of
an inhibition rule formed only by one inhibiting property or a group of inhibiting properties, with a weight of
inhibition, and associating it to the decision rule or to the group of rules to inhibit. The concept of decisional
inhibition is represented by a circle including the symbol I- and associated with an arrow from the inhibitor rule
towards the inhibited rule or grouping of rules.
 Concept of decisional multiplicity: This means authorizes the optional and/or multiple participation, with
different weights, of a decision rule in the computing of the d.o.m of an entity e in a class. The concept of
decisional multiplicity is graphically represented by a set of weights or if all the weights have the same value, we
can simply add the symbol * before the value of the weight. If this rule is optional, its weight or its set of weights
is written between hooks.
 Concept of decisional selection: This concept makes possible the selection among several ones and
according to a given criterion (Max, Min), a decision rule or a grouping of decision rules. The concept is
graphically represented by a discontinued arc comporting a selection criterion.
The enrichment of FSM also concerns the parameterization of decision rules in calculating the d.o.m of
entities [5]; a decision rule relative to a class can be based on attributes belonging to other classes, on objects of
different other classes and on total semantics. Thus, a rule is not only attached to its class (for which it is
involved in calculating the degree of membership) with a certain weight, but it can also be attached to any other
class comprising the attribute or the object on which it is based by a line segment containing in its middle a small
square which can be shared by several decision rules referring to the same class.
An illustration of the concepts of EFSM on a practical case is given through the example treating the
disease entitled Multiple Sclerosis “MS” in Figure 1. This model is composed of five strong classes (Patient,
Diagnostic_type, Symptom and MS) and two weak ones (Noticed_symptom and Likely_symptom). Thirteen
decision rules, of which ten are object-based and three are attribute-based, are related to MS class. However, they
are also attached to the classes containing the object or the attribute on which they were based. Two among these
rules are optional (whose weight is written in square bracket) and one is optional and multiple at the same time
(the weight is between hooks and brackets).
The model also comprises five nodes of decisional grouping whereof each one regroups a subset of these
rules. The node n5 is in turn a node whose participation in the calculation of the d.o.m in MS is optional but of
which the degree of membership of an entity e to itself is the max between two degrees of membership relating
to the nodes n3 and n4; we talk here about the decisional selection. The concept of decisional inhibition is also
present in the model with the inhibiting rule Presence of febrile situation and the inhibiting node n2.
To be able to apply these concepts, we need a framework for modeling and managing fuzzy semantic
databases based on EFSM. One of the principal functions of this framework is to allow the representation of a
model in accordance with EFSM concepts. So, we focuse in a first step on defining a graphical editor by
adopting the MDA approach.
19
4. Development of a graphical editor for EFSM according to MDA approach

To obtain an editor who satisfies our requirements, it is necessary to go through successive phases. The
preliminary phase concerns the census of needs specifying the objectives to be reached and the constraints to be
ensured. After this phase, we obtain a model at the business level which will serve as specifications in the design
phase. This latter produces needed models for the implementation phase.
In fact, treating a problem throw phases simplifies its resolution, i.e., reduces the difficulty of automating the
application by focusing on each phase in the study of appropriate solutions that are in a level of abstraction
higher than the phase that follows. The successive refinement of the level of abstraction allows mastering the
complexity of development.
Figure 1: Extract of the diagram relative to MS disease under EFSM [5]
This strategy of work is supported by the MDA approach whose principal key consists of the use of models at
different phases of the development cycle of an application. MDA recommends initially the elaboration of
requirements models (CIM: Computational Independent Model), then analysis and design models (PIM:
Platform Independent Model) and finally the code (PSM: Platform Specific Model). Its main objective is to
reduce the technological complexity of handling the problem through different models separating the functional
specifications from those of its implementation on a given platform. Furthermore, MDA enables saving time,
better portability and interoperability of models and an easier maintenance of the software.
Let see now our productions according to the hierarchy of MDA.
1.1. The requirement model of EFSM-Editor
At the CIM level, we express our needs towards the manager of models that we want to develop. In fact,
this manager should essentially enable us to handle EFSM-based models, according to the concepts,
relationships between concepts, graphical representations of constructs and constraints defined in [4, 5, 19].
Our model of requirements can be summarized with the use case diagram in Figure 2. Indeed, this diagram
contains the functionalities to be provided by the manager of models as well as the entity that interacts with it
(the user in our case). Precisely, the designer for EFSM models uses the editor for model management including
creation, modification and deletion.
Our model of business level can be presented using a class diagram representing a meta-model although the
latter has a rather static aspect. Our starting point is the EFSM meta-model defined under UML. In this latter, the
20
majority of classes represent specializations of abstract classes Class, Relationship, Attribute and Decisional
Concept which represent the central elements in the meta-model. This is why we gathered them in an abstract
class named Element. This class collects the components of any EFSM model which is indicated using the class
Model. For each element in an EFSM model corresponds a graphical symbol to which we granted a class named
Figurine which also presents a component class of Model containing all the graphic forms: nodes and
connections. The obtained result is shown through Figure 3. This model allows defining traceability links with
models built in future phases.
A designer for
EFSM models
User <<include>>
Management of an EFSM Management of an EFSM

model element
Definition of a decisional
concept
Definition of a class
Definition of an attribute
Definition of a relationship Definition of a link
Figure 2: Use case diagram modeling the requirements of EFSM-Editor
1.2. The computational models of EFSM-Editor

We must proceed at this level to the development of models describing the system, but without showing the
details of its implementation on a particular platform. These models should be elaborated from the CIM. It is a
computer representation of the requirements through a kind of transformation of models (CIM to PIM), which is
generally done manually.
1.2.1. PIM1 of EFSM-Editor
The formalism that we use here to express the first PIM is an UML class diagram representing a meta-
model, coupled with the object constraint language OCL. This model has the same root class Model as the meta-
model of the CIM level (cf. Figure 3). However, a restructuring of this latter is needed to provide a vision
focused on the computational side, with distinguishing between different EFSM concepts by the type of
graphical representation. All EFSM elements inherit now from Node class or Link class.
This reorganization returns to the fact that the essential functionality of the editor consists in offering the
possibility to represent each concept of EFSM using a figurine. This model (cf. Figure 4) is mainly formed by
two branches modeled using two compositions of the root class: the first collects together elements of the model
which will be represented by a figurine in the form of a node (rectangle, rounded rectangle, circle, …), such as
different types of classes, attributes, relationships that may exist between classes and the different decisional
elements having specific figurines; the other, named Link, represents all links (Line or Arrow) between these
nodes. The class Link replaces all connections between the various nodes of the model.
This model mainly enables attributing to elements, having specific symbols in EFSM, figurines in
conformity with these symbols. However, in the passage from the CIM towards this model (PIM1), we note that
several concepts will be represented by the same type of figurine. For that, we have identified two principal rules
of transformation to ensure the passage between these two levels:
R1: Every concept having a specific figurine is represented by a class.
21
Figure 3: The CIM of EFSM-Editor
Figure 4: First PIM of EFSM-Editor
R2: All the concepts having the same type of figurine are represented by the same class, with specifying, if
necessary, a set of OCL constraints to complete semantics. Among these constraints, we mention the ones
defining the authorized links between the various figurines and the ones specifying the multiplicities of all links
22
(or number of links between each two types of figurines).

Hereafter we give some examples of applying these two rules on our CIM:
By applying R1:
 The class Attribute-based decision rule will appear in the PIM because it has a specific figurine.
 The attribute Selection criterion in the Decisional selection class is transformed into two separate classes,
each one has a specific figurine representing the appropriate criterion (MAX or MIN).
By applying R2:
 All links existing between the classes of the CIM are to gather in the same class Link. They can be lines or
arrows. To illustrate the semantics of these links, let’s take two examples of application. These are the two
constraints CT1 and CT2 listed below.
CT1: The node Gen/Spec relationship roster-defined can be connected only to Simple classes.
CT2: The node Gen/Spec relationship roster-defined can be connected to one or more Simple classes as sub-
class and to one as super-class.
 The fuzzy and crisp simple classes are collected in the same class, which is the class Simple class, since they
have the same figurine (a simple rectangle). Similarly, fuzzy and crisp composite classes are gathered in the class
Composite class.
1.2.2. PIM2 of EFSM-Editor
Since we cannot transform a PIM into a PSM only when it is sufficiently refined, we proceeded to an
enrichment of the previous model by information specifying complementary aspects, in two steps:
 1st step: It consists of reducing the number of OCL constraints by substituting them by simple interaction
links between different nodes.
It is a question of fixing the incoming and outgoing links for each node in the model with specifying the
multiplicity (non showed in Figure 5 to ensure its readability) and the opposite link of each one. Moreover, in the
case where links have specific attributes, they are to be transformed into specialized classes such as Branch,
Decisional link and Inhibiting link (cf. Figure 5).
This type of transformation enables us to avoid a set of constraints for the majority of the classes of PIM1
while keeping the semantics in the model.
Example:
CT1 and CT2 constraints, which concern the node Gen/Spec relationship roster-defined, can be replaced by an
incoming link and an outgoing link from that node to the node Simple class, accompanied by the multiplicity and
the opposite link of each one to well ensure these constraints. Same links are to be defined for the class Gen/Spec
relationship attribute-defined.
Figure 5: Second PIM of EFSM-Editor
23
 2nd step: It consists in decreasing the number of these links by the addition of generalization classes while
trying to keep the semantics of the model. These new classes allow making groupings ensuring a kind of
factorization of several links which are repeated in different nodes.
Example 1: Any type of decision rule (Object-based decision rule, Attribute-based decision rule or Semantic
decision rule) has interaction links with the class Connecting node. We added thus a super-class for those three
types of rules in order to factor these links.
Example 2: We've created a new class called Other, as a generalization class of Gen/Spec relationship roster-
defined and Gen/Spec relationship attribute-defined classes. This reduces the number of links by replacing those
of the two nodes by links in the new class.
This refinement of model gave birth to the model in Figure 5. This model has the same organization of classes of
the PIM1 but with additional classes occupying the position of sub-classes (derived from the first step of
refinement) for the Line and Arrow classes or the position of super-class, each one for a subset of sub-classes of
the Node class in PIM1 (derived from the second step of refinement). All these new classes are marked in gray in
Figure 5.
1.3. The code models of EFSM-Editor
According to [20], a PSM is a sight of the system viewed from the perspective of the platform
implementation: it covers the technical level of the architecture. PSM combines the specifications of the PIM
with the characteristics that specify how the system can function on the target platform.
In this phase, we propose a solution to move from an abstraction to an implementation: the transition from PIM
to PSM constitutes an important aspect of the MDA approach.
The code model will be derived from PIM2 with taking into consideration the platform in which our editor
will be developed. A multitude of tools for the implementation of the MDA approach can satisfy our needs, the
most advanced are:
- DSL Tools of Microsoft [DSL_Tools] (Visual Studio Extensibility),
- Eclipse EMF/GEF/GMF platform [Eclipse Project],
- MDA Modeler – Objecteering UML Modeler.
The comparison of these platforms made by [21] showed that the solution Eclipse EMF / GEF / GMF offers
more flexibility in terms of graphical representation. It is considered as the reference platform to implement the
approach based on models. We then adopt this platform to increase the graphics capabilities of our tool.
Our choice is fixed on the GMF plugin that combines the two others (EMF and GEF) and makes it possible to
create graphical editors in accordance with a given model. This plugin allows the automatic transformations
PIM-PSM and automatic generation of the code. Like any model transformation tool, it takes as input a model
and transforms it into another model. At the current level of MDA, the entry model is often the PIM and the
target model is often the PSM.
Figure 6: Screenshot of EFSM-Editor
24
Applied to our work context, the implementation of the graphical editor through GMF was structured as
follows:
 We start by providing the PIM2 as a domain model corresponding to our application through the ecore
model.
GMF allows making of chain of transformations from this model to generate a graphical interface. Each
transformation is represented through a model.
 gmfgraph model: This model is interested in the description of the figurines of representation (rectangle,
ellipse,…).
 gmftool model: We define here the elements that will appear in the editor palette as well as their icons.
 gmfmap model: At this step, a description of the correspondences between the concepts of the meta-model,
the figurines and the tools is made to accord to each figurine the ecore element that it will represent and its
corresponding icon in the palette
Finally, to generate the editor prototype, the gmfgen model, which synthesizes the information issue from
these different models, is generated. And from the latter, the editor diagram is produced according to what has
been defined through different transformations.
GMF is responsible for all these transformations from the ecore to the development of the editor
automatically by using simple clicks. However, manual intervention is necessary in the elaboration of different
models to customize the editor according to our needs. This editor is shown through Figure 6. The editor
interface is divided mainly into five blocks: a toolbar, a central bloc for diagram representation, a left bloc
containing the instance of the model as a tree, a bloc in a rectangle form allowing the visualization of properties
of a selected element from the model, and finally a small square showing an overview of the model.
Since XMI (XML Metadata Interchange) is in the second ring of the standards used by MDA [22], and to
ensure the persistence property, EFSM-Editor permits to save each diagram in two separate XMI resources: the
main file (which contains an instance of the EFSM meta-model, i.e., contains every data having a semantic
aspect concerning all elements represented in the model) (cf. Figure 7) and the diagram file (which contains the
notations of the graphic elements).
In fact, the first file was the entry of a code generator that we developed using the OAW (Open Architecture
Ware) tool with the Xpand template language. The code is generated from any model created with EFSM-Editor
in accordance with the definition language for fuzzy classes’ schema, defined in [3]. However, we have modified
this language in its EXTENT clause (which concerns the definition of decision rules for the relevant class) for
taking into account the new concepts introduced in [4, 5].
5. Conclusion
After the introduction of new concepts [4, 5], FSM can better present the real word. However, the absence
of a tool CASE to model fuzzy semantic data was an obstacle.
This paper aimed to define and implement a framework of graphical editor for the representation of fuzzy
semantic models based on EFSM. The proposed solution takes into account the new concepts introduced in the
fuzzy semantic model and follows the architecture of models MDA.
This paper discusses in details the various phases of this approach in accordance with our work context by
specifying the passages between models of various steps. In fact, our initial model was the EFSM meta-model
which we used for defining models of different levels (CIM / PIM / PSM) of the MDA approach.
Moreover, the OCL language was the most suitable language for defining constraints in the description of
transformation rules between MDA models. However, the task of implementing these constraints is still not
completed. We are also looking for a refinement for the editor by improving its features for more interactivity
with the user.
6. Bibliography
[1] Ma Z. M. and Yan Li., (2008), “A Literature Overview of Fuzzy Database Models”, Journal of information
science and engineering, Vol. 24, pp. 189-202.
[2] Ram S., “Intelligent database design using unifying semantic model”, (1995), Information and Management,
Vol. 29, No. 4, pp. 191–206.
[3] Bouaziz R., Chakhar S., Mousseau V., Ram S. and Telmoudi A., (2007), “Database design and querying
within the fuzzy semantic model”, Information Sciences, Vol. 177, Issue 21, pp. 4598-4620.
[4] Ellouze L., Bouaziz R. and Chakhar S., (2009), “Extending Fuzzy Semantic Model by Advanced Decision
Rules”, Fuzzy Information Processing Society, Annual Meeting of the North American, NAFIPS,
Cincinnati, OH.

The ecore format is the Eclipse implementation of the MOF specification.
25
[5] Ellouze L., Bouaziz R. and Chakhar S., (2009), “Nouveaux concepts pour mieux modéliser le flou”, Actes
du XXVII° congrès INFORSID, Toulouse, pp. 177-192.
[6] Ma Z. M., Zhang W. J. and Ma W. Y., (2000), “Semantic measure of fuzzy data in extended possibility-
based fuzzy relational databases”, International Journal of Intelligent Systems, Vol. 15, pp. 705-716.
[7] Prade H. and Testemale C., (1984), “Generalizing database relational algebra for the treatment of
incomplete or uncertain information,” Information Sciences, Vol. 34, pp. 115-143.
[8] Raju K. V. S. V. N. and Majumdar A. K., (1988), “Fuzzy functional dependencies and lossless join
decomposition of fuzzy relational database systems”, ACM Transactions on Database Systems, Vol. 13, pp.
129-166.
[9] Cross V., Caluwe R. and Vangyseghem N., (1997), “A perspective from the fuzzy object data management
group (FODMG)”, Proceedings of the 1997 IEEE International Conference on Fuzzy Systems, Vol. 2,
Barcalona, Catalonia, Spain, pp. 721-728.
[10] Cross V. and Firat A., (2000), “Fuzzy objects for geographical information systems”, Fuzzy Sets Systems,
Vol. 113, pp. 19-36.
[11] Shukla P. K., Tripathi S. P. and Bajpai S. K., (2010), “Handling grey information in object oriented
databases”, International Journal of Information Technology and Knowledge Management, Vol. 2, No. 2,
pp. 211-215.
[12] Chen G. Q., (1998), “Fuzzy logic in data modeling, semantics constraints, and databases design”. In A. K.
Elmagarmid (Ed.), The Kluwer international series on advances in database systems.
[13] Ma Z. M., Zhang W. J., Ma W. Y., and Chen Q., (2001), “Conceptual design of fuzzy object-oriented
databases using extended entity-relationship model”, International Journal of Intelligent Systems, Vol. 16,
No. 6, pp. 697-711.
[14] Yazici A., Buckles B.P. and Petry F.E., (1999), “Handling complex and uncertain information in the ExIFO
and NF2 data models”, IEEE Transactions on Fuzzy Systems, Vol. 7, No. 6, pp. 659-676.
[15] Yazici A. and Cinar A., (2000), “Conceptual modeling for the design of fuzzy object oriented database”, In
O. Pons, M. A. Vila, & J. Kacprzyk (Eds.), Knowledge management in fuzzy databases, Physica-Verlag,
pp. 12-35.
[16] Zvieli A. and Chen P., (1986), “ER modeling and fuzzy databases”, In Proceedings of the Second
International Conference on Data Engineering, pp. 320-327.
[17] Chen G. Q. and Kerre E. E., (1998), “Extending ER/EER concepts towards fuzzy conceptual data
modeling”. In Proceedings of the IEEE International Conference on Fuzzy Systems, Vol. 2, pp. 1320-1325.
[18] Vila M.A., Cubero J.C., Medina J.M. and Pons O., (1996), “A conceptual approach for deal with
imprecision and uncertainty in object-based data models”, Int. J. Intell. Systems, Vol. 11, pp. 791-806.
[19] Bouaziz R., Chakhar S. and Saad I., (2005) “Membership functions definition in the fuzzy semantic
model”, International Conference: Sciences of Electronic, Technologies of Information and
Telecommunications (SETIT'05), Sousse, Tunisia.
[20] Miller J. and Mukerji J., (1995). “MDA Guide Version 1.0.1”, Web document [OMG-MDA 2003].
http://www.omg.org/ docs/omg/03-06-01.pdf.
[21] TURKI S., (2008), “Ingénierie système guidéepar les modèles: Application du standard IEEE 15288, de
l’architecture MDA et du langage SysML à la conception des systèmes mécatroniques. Sciences et
techniques industrielles”, DEA MCAO Aix-Marseille III : L’Université du Sud Toulon-Var.
[22] Kadima H., (2005), “MDA Conception orientée objet guidée par les modèles”, DUNOD, Etudes &
Développement, PARIS.
26
Figure 7: Extract of XMI code generated relative to SEP model [5]
27
International Journal of Computational Intelligence and Information Security, Vol. 1, No. 5, July 2010
Credit Card Fraud Detection using Kernel in SVM

Harish Baraithiya*, R.K.Pateriya**
*
M.Tech (Information Security) Scholar, Maulana Azad National Institute of Technology,
**
harish.baraithiya@gmail.com*, pateriyark@gmail.com**
Abstract
The growth of ecommerce increases the money transaction via electronic network which is designed for
hassle free fast & easy money transaction. The facility involves greater risk of misuse for fraud, one of them is
credit card fraud which can happen by many types as by stolen card, by internet hackers who can hack your
system & get important information about your card or by information leakage during the transaction. Several
researchers have proposed their work for credit card fraud detection by characterizing the user spending profile.
In this paper we are proposing the SVM (Support Vector Machine) based method with multiple kernel
involvement which also includes several fields of user profile instead of only spending profile. The simulation
result shows improvement in TP (true positive), TN (true negative) rate, & also decreases the FP (false positive)
& FN (false negative) rate.
Keywords: Fraud detection, SVM (support vector machine), kernels.
28
1. Introduction
Growth in communication network, increased internet speed, easy wireless connectivity & lack of time
causes the people to buy through electronic network. Here are some statistics and projections of the Indian credit
card industry (http://hubpages.com/hub/Indian-Credit-card-Industry) to show importance of the topic.
1. India is currently the fastest growing Mobile Market in the world and is also among the fastest growing credit
card markets in the world.
2. India has a total approx.75 million cards under circulation (25 million credit and 50 million debit) and a 30%
year-on-year growth.
3. With 87% of all transactions in plastic money happening through credit cards, debit cards in India continue
to be used largely for cash withdrawals.
4. Though Visa, which accounts for 70% of the total card industry is the market leader in India; MasterCard is
fast catching up.
5. Every transaction involves payment of an interchange charge to MasterCard or Visa for settlement, which
amounted to about $50 million during the year.
6. Internal estimates of Barclaycard have pegged the Indian market with potential to grow to at least 55million
credit cards by 2010-11.
The above statistics shows the money involved in transaction through cards & it is required to insure the security
of money for both the Bank & for customer.
2. Related Work
As we stated before that many researchers have proposed their work in this area out of which the most
relevant to our work are, the work done by Abhinav Srivastava, Amlan Kundu, Shamik Sural, Arun K.
Majumdar [1] have proposed the probabilistic model based on HMM(Hidden Markov Model) they consider the
spending history of card holder & characterize the spending pattern by dividing the transaction amount in three
category shows the TP rate of 0.65 & FP rate of 0.05. Another relevant contribution is by Wen-Fang YU & Na
Wang [2]. They have proposed the distance based method. This method judge whether it is outlier or not
according to the nearest neighbors of data objects. They only showed the highest accuracy about 89.4 percent but
not talked about FP & FN. A neural network based approach is presented by Sushmito Ghosh and Douglas L.
Reilly [3] in their paper they selected large set of 50 field & after proper relation it is reduced to set of 20
features which is used for training neural network. The neural network used in this fraud detection feasibility
study is the P-RCE neural network. The P-RCE is a member of the family of radial-basis function networks that
have been developed for application to pattem recognition. The P-RCE is a three-layer, feed-forward network
that is distinguished by its use of only two training passes through the data set. Same work is also done by using
regression techniques & compared against neural & decision tree methods [4] this work is done by Aihua Shen,
Rencheng Tong, Yaochen Deng.their simulation shows that neural networks model provides higher lift(Lift table
and lift chart were used to describe the usefulness of the model to create the scored data set. "Lift" is probably
the most commonly used metric to measure the performance of targeting models in classification applications.)
than a logistic regression and decision tree on the
same data, while neural networks slightly better than logistic regression. This provides a key factor in choosing
the models. A similar coefficient sum based model analysis explained by Chun Hua Ju & Na wang [5] they
analyze type I & type II error rate with highest rate of TP up to 89 percent.
3. SVM (Support Vector Machine)
Support Vector Machines (SVMs) have been developed from Statistical Learning Theory [6]. The SVM are
widely applied to fields such as character, handwriting digit and text recognition, and more recently to satellite
image classification. SVMs, like ANN and other nonparametric classifiers have a reputation for being robust.
SVMs function by nonlinearly projecting the training data in the input space to a feature space of higher
dimension by use of a kernel function. This results in a linearly separable dataset that can be separated by a
linear classifier. This process enables the classification of datasets which are usually nonlinearly separable in the
input space. The functions used to project the data from input space to feature space are called kernels (or kernel
machines), examples of which include polynomial, Gaussian (more commonly referred to as radial basis
functions) and quadratic functions. Each function has unique parameters which have to be determined prior to
classification and they are also usually determined through a cross validation process. A deeper mathematical
treatise of SVMs can be found in [7].
By their nature SVMs are intrinsically binary classifiers however there exist strategies by which they can be
adapted to multiclass tasks. But in our case we not need multiclass classification.
29
3.1 SVM classification

Let xi  Rm be a feature vector or a set of input variables and let yi  {+1, -1} be a corresponding class label,
where m is the dimension of the feature vector. In linearly separable cases a separating hyperplane satisfies [8].
Where the hyperplane is denoted by a vector of weights w and a bias term b. The optimal separating hyperplane,
when classes have equal loss-functions, maximizes the margin between the hyperplane and the closest samples
of classes. The margin is given by
The optimal separating hyperplane can now be solved by maximizing (3) subject to (1). The solution can be
found using the method of Lagrange multipliers. The objective is now to minimize the Lagrangian
and requires that the partial derivatives of w and b be zero. In (4), i are nonnegative Lagrange multipliers.
Partial derivatives propagate to constraints w =i iyixi and iiyi = 0. Substituting w into (4) gives the dual
form
which is not anymore an explicit function of w or b. The optimal hyper plane can be found by maximizing (5)
subject to iiyi = 0 and all Lagrange multipliers are nonnegative. However, in most real world situations classes
are not linearly separable and it is not possible to find a linear hyper plane that would satisfy (1) for all i = 1. . .
n. In these cases a classification problem can be made linearly separable
By using a nonlinear mapping into the feature space where classes are linearly separable. The condition for
perfect classification can now be written as
where  is the mapping into the feature space. Note that the feature mapping may change the dimension of the
feature vector. The problem now is how to find a suitable mapping  to the space where classes are linearly
separable. It turns out that it is not required to know the mapping explicitly as can be seen by writing (6) in the
dual form
and replacing the inner product in (7) with a suitable kernel function K(xj , xi) = ((xj) · (xi)). This form arises
from the same procedure as was done in the linearly separable case that is, writing the Lagrangian of (6), solving
partial derivatives, and substituting them back into the Lagrangian. Using a kernel trick, we can remove the
explicit calculation of the mapping  and need to only solve the Lagrangian (5) in dual form, where the inner
product (xi · xj) has been transposed with the kernel function in nonlinearly separable cases. In the solution of
the Lagrangian, all data points with nonzero (and nonnegative) Lagrange multipliers are called support vectors
(SV).
Often the hyper plane that separates the training data perfectly would be very complex and would not generalize
well to external data since data generally includes some noise and outliers. Therefore, we should allow some
violation in (1) and (6). This is done with the nonnegative slack variable i
The slack variable is adjusted by the regularization constant C, which determines the tradeoff between
complexity and the generalization properties of the classifier. This limits the Lagrange multipliers in the dual
objective function (5) to the range 0  i  C. Any function that is derived from mappings to the feature space
satisfies the conditions for the kernel function.
The choice of a Kernel depends on the problem at hand because it depends on what we are trying to model.
A polynomial kernel, for example, allows us to model feature conjunctions up to the order of the polynomial.
30
Radial basis functions allows to pick out circles (or hyper spheres) - in contrast with the Linear kernel, which
allows only to pick out lines (or hyper planes).
Linear Kernel: The Linear kernel is the simplest kernel function. It is given by the inner product <x,y> plus an
optional constant c. Kernel algorithms using a linear kernel are often equivalent to their non-kernel counterparts,
i.e. KPCA with linear kernel is the same as standard PCA.
Polynomial Kernel: The Polynomial kernel is a non-stationary kernel. Polynomial kernels are well suited for
problems where all the training data is normalized.
Adjustable parameters are the slope alpha, the constant term c and the polynomial degree d.
Gaussian Kernel: The Gaussian kernel is an example of radial basis function kernel.
Alternatively, it could also be implemented using
The adjustable parameter sigma plays a major role in the performance of the kernel, and should be carefully
tuned to the problem at hand. If overestimated, the exponential will behave almost linearly and the higher-
dimensional projection will start to lose its non-linear power. In the other hand, if underestimated, the function
will lack regularization and the decision boundary will be highly sensitive to noisy training data.
The SVM gives the following advantages over neural networks or other AI methods (link for more details
http://www.svms.org).
SVM training always finds a global minimum, and their simple geometric interpretation provides fertile ground
for further investigation.
Most often Gaussian kernels are used, when the resulted SVM corresponds to an RBF network with Gaussian
radial basis functions. As the SVM approach “automatically” solves the network complexity problem, the size of
the hidden layer is obtained as the result of the QP procedure. Hidden neurons and support vectors correspond to
each other, so the center problems of the RBF network is also solved, as the support vectors serve as the basis
function centers.
Classical learning systems like neural networks suffer from their theoretical weakness, e.g. back-propagation
usually converges only to locally optimal solutions. Here SVMs can provide a significant improvement.
The absence of local minima from the above algorithms marks a major departure from traditional systems such
as neural networks.
SVMs have been developed in the reverse order to the development of neural networks (NNs). SVMs evolved
from the sound theory to the implementation and experiments, while the NNs followed more heuristic path, from
applications and extensive experimentation to the theory.
4. Proposed works
4.1 Proposed Algorithm
Here we detail the proposed algorithm for classification of Fraud Transactions.
Step 1: Read the given data.
Step2: Re-categorize the data into five groups as transaction month, date, day, amount of transaction &
difference between successive transaction amounts.
Step 3: Make each transaction data as vector of five fields.
Step 4: Make two separate groups of data named True & False transaction group (if false transaction data is not
available add randomly generate data in this group).
Step 5: Select one of three kernels (Linear, Quadratic, and RBF).
31
Step 6: Train SVM.

Step 7: Save the classifier.
Step 8: Read the current Transaction.
Step 9: Repeat the process from step1 to step3 for current transaction data only.
Step10: Place the saved classifier & currently generated vector in classifier.
Step 11: Take the generated decision from the classifier.
4.2 Proposed model for fraud detection
Input transaction probability Select SVM kernel
Generated synthetic data

Linear Quadratic RBF
Input false transaction

Labeledprobability
data
Categorized data Generated classification

equation
Current transaction data Decision equation
Output decision
Figure 1: Fraud detection model

5. Experiments and Results
Since there is no real data is available because of privacy maintained by banks. Hence for testing &
implementation of our algorithm we have generated the data of true & false transaction using different mean &
variance & then mixed them with different probability. We have used MATLAB for the implementation of
algorithm because of its rich sets of mathematical function and also supporting the inbuilt functions for SVM.
We have simulated our algorithm for five different fraud probabilities from 0.3 to 0.5 & changing the training
data size from 30 to 100, and calculated
TPR = True Positive Rate
TNR = True Negative Rate
FPR = False Positive Rate
FNR = False Negative Rate.
A comparison between Kernel Type : Linear, Quadratic and RBF was performed which shows that the RBF
kernel outperform to Linear & Quadratic kernel. It has maximum accuracy up to 97%, maximum
TPR(99%),maximum TNR(98%) & maximum FPR(7%),maximum FNR(6%), it also behaves almost same for
all type of data set generated(having very low fraud data & high fraud data)
7. Conclusion and Future Scope
Referring to results we can say that proposed algorithm with RBF kernel gives the better results in comparison
with the previous work we have discussed & hence can be used for automatic Credit card Fraud detection with
excellent accuracy & minimum false alarm.
We can enhance this model for dynamic improvements in training of classifiers using different SVM models like
incremental SVM detrimental SVM, evolutionary SVM etc. but we leave this job for future.
32
Acknowledgments
researchers.
References
[1] Abhinav Srivastava, Amlan Kundu, Shamik Sural, Arun Majumdar, "Credit Card Fraud Detection Using
Hidden Markov Model," IEEE Transactions on Dependable and Secure Computing, vol. 5, no. 1, pp. 37-
48,Jan.-Mar 2008, doi:10.1109/TDSC.2007.70228
[2] Wen-Fang Yu, Na Wang, "Research on Credit Card Fraud Detection Model Based on Distance Sum," jcai,
pp.353-356, 2009 International Joint Conference on Artificial Intelligence, 2009
[3] Sushmito Ghosh, Douglas L. Reilly. “Credit Card Fraud Detection with a Neural-Network.” In: Proceedings
of the 27th Hawaii International Conference on Information Systems, S. 621-- 630, 1994.
[4] Aihua Shen, Rencheng Tong, Yaochen Deng. “Application of Classification Models on Credit Card Fraud
Detection.” In: Proceedings of the International Conference on Service Systems and Service Management,
pp.1-4, 2007.
[5] ChunHua Ju, Na Wang, "Research on Credit Card Fraud Detection Model Based on Similar Coefficient
Sum," dbta, pp.295-298, 2009 First International Workshop on Database Technology and Applications,
2009.
[6] Vladimir N. Vapnik, The nature of statistical learning theory, Springer-Verlag New York, Inc., New York,
New York: Springer-Verlag, 1995.
[7] V. Kecman, Learning and Soft Computing: Support Vector Machines,. Neutral Networks and Fuzzy Logic
Models. Cambridge, MA: MIT. Press, 2001.
[8] Seppo Fagerlund, “Bird species recognition using support vector machines,” EURASIP Journal on Applied
Signal Processing, v.2007 n.1, p.64-64, 1 January 2007.
33
Attribute vector optimization for a non supervised classification by

evolutionary approach
F. Lekhal1, M. El Hitmy1, O. El Melhaoui1 and M. Nasri2
1
LABO LEPAS, FS, University Of Mohammed I, OUJDA, MOROCCO
2
LABO MATSI, EST, University Of Mohammed I, OUJDA, MOROCCO
lekhal_ f@yahoo.fr
Abstract
The classification requires the selection of a set of discriminate attributes for describing the data. We
present in this work a new algorithm for optimizing the attribute vector in a non supervised classification. The
algorithm begins by selecting the optimal number of classes based on the initially chosen set of attributes, the
Xie and Beni criterion and the EFCM algorithm. The EFCM is based on a fuzzy C-means FCM and an
evolutionary strategy approach. The next step is to optimize the attribute vector by minimizing a cost function
and by using a genetic algorithm. The cost function used by the genetic algorithm is based on a new criterion
which we have recently proposed and which is built on an AURAY criterion and a correlation approach. The
proposed algorithm is validated on two types of data which are the texture images and alphabetic letters.
Keywords: Classification, evolutionary strategies, fuzzy C-means, evolutionary fuzzy C-means, Xie and Beni
criterion, genetic algorithm, AURAY criterion, selection of parameters.
34
1. Introduction
Data classification consists of regrouping the data into different classes. The data of a same class are all similar
according to some criterion. The data of two different classes are dissimilar according to the same criterion. Data
classification has played an important role for resolving problems in pattern recognition, imaging, image
segmentation and has been used in fields such as medicine, biology, industry, telecommunications etc.
Data classification is of unsupervised type if the prior information about the data is entirely or partly not
available. It is of supervised type if some necessary a priori information about the data is given [1, 2, 14]. Several
methods are known and used for data classification [1]. We use in this work, for data classification, the fuzzy C-
means method which is based on the C-means and the fuzzy set theory. We present in this work an evolutionary
adaptation to the FCM [10].
The attribute parameters used to describe the data are either qualitative, quantitative, structured or statistical [1].
In order to distinguish in a clear way, between two different objects, it is highly desirable to select the most pertinent
attribute parameters. The optimal choice of these parameters gives rise to a rapid and concise decision and may play,
for the classification, the role of a filter against the noise brought about by the non representative parameters [1, 11].
We begin, in this work, by selecting the optimal number of classes Copt, by means of the initially chosen set of
attributes. We perform the task of data attribution; each data is set to its specific class. We then optimize the attribute
vector using a genetic approach.
2. Data description
2.1. Texture parameters
Agriculture fields, wood, soil, clothing material, leather, wall etc, are all examples of texture images. The
classification of these types of images is based on the representation of the images in an attribute space. Several
methods of characterizing texture images were proposed in the literature [18]. Haralick [12] has proposed fourteen
parameters which were extracted from a co-occurrence matrix. The most frequently used parameters are: Hom,
HomL, Uni, Dir, Ent and Cont. Galloway [13] has proposed five parameters which were extracted from a run length
matrix. These parameters are: SRE, LRE, GLN, RLN and RP.
Vinit = (Dir, Uni, Hom, Cont, HomL, Ent, SRE, LRE, GLN, RLN, RP)
2.2. Alphabetic letters parameters
Recognizing a letter requires the analysis of its shape, and the extraction of its special characteristics which will
be exploited for its identification. The characteristic types may be divided into four principal groups [19], structural,
statistical, global transformations, and the superposition of models and correlation. In this work we are only
interested by the statistical parameters. Eleven parameters are commonly used in this case [20], these are : height-
width rate (R), the number of pixels used for the letter image (PN), the gravity center coordinates (Gx, Gy), the letter
pixels density (D), the number of connected Components, in 4-connexity and 8-connexity (Cn4, Cn8), the three
inertia moments, in x, y and xy (Ix, Iy, Ixy) and the orientation angle (OA).
Vinit = (R, PN, Gx, Gy, D, Cn4, Cn8, Ix, Iy, Ixy, OA)
3. Evolutionary algorithms
Evolutionary algorithms have proved to be very efficient stochastic optimization algorithms of functions
[15,18]. These techniques are based on the evolution of a population of solutions, which is updated at each iteration,
called new generation. The population size is maintained constant at each step. For updating the population, the
algorithm uses some precise rules for optimizing a given behavior which is expressed by a fitness function [16]. The
evolutionary algorithm sets out the value of the initial population elements. At each iteration, a new population is
obtained by the application of genetic operators: selection, crossover and mutation in GA and selection and mutation
in SE case. These operators are stochastic. Figure 1 and 2 show the different operations undertaken, respectively in a
basic genetic algorithm [16] and in basic evolutionary strategies algorithm [15].
35
Random generation of an initial

population
Compute the fitness function
Repeat
1- Selection
2- Crossover
3- Mutation
4- Compute the fitness function
Figure 1: Basic genetic algorithm.
Random generation of an initial population

Compute the fitness function
Repeat
1- Choose the parents
2- Modify the generation by mutation (by
using Gaussian perturbation)
3- Select the new generation
4- Compute the selective function
Until the stop criterion is reached
Figure 2: Standard SE algorithm.
4. Fuzzy classification
4.1. Descriptive elements
Let’s consider a set of M objects { Oi }, characterized by N parameters regrouped in a line vector V = (aj). Let Ri
= (aij)1≤j≤N be a line vector of RN where the jth component aij is the value taken by aj for the objet Oi. Ri is called the
observation associated with Oi, it is also called the realization of the attribute vector for this object. RN is the
observation space, it is also known as the parameters space. Let EV = {a1, a2, .., aj, .., aN} be the set of attributes
associated to V.
4.2. Fuzzy C-means algorithm.
The underlying principle of the FCM [2,3,4] is to form from M objects defined by their realizations (Ri)1≤i≤M, C
classes (CLs)1≤s≤C. The classes must be highly homogenous. This is expressed mathematically by searching for the
class centers (gs)1≤s≤C, which minimize the optimization criterion defined by [2,4], and the constraints:
M C df
J m    is 
2
Ri  g s
i 1 s 1
Under the constraints:
C M
  is  1 for i=1 to M and 0 <  is < M for s=1 to C
s 1
i 1
||.|| is the Euclidean distance, μis is the membership degree of the object i to the class CLs. all the μis belong to
the interval [0,1] and from a matrix of size M × C, called membership matrix μ. df is the ‘’fuzzy degree’’ often, df is
equal to 2. In order to minimize Jm, μis and gs , must be updated at each iteration according to [3,18]:
1
 R  g 2 1 df
 i s 
is   
1
C  2 1 df
  Ri  g k 
k 1 
and
36
 
i 1
is df Ri
gs  M
 
i 1
is df
Figure 3 shows the different operations undertaken in an FCM algorithm [3, 18].
Fix the number of classes C

Initialize randomly the centers
Stop = false
While Stop = false Do
Adjust the degrees of membership
Adjust the centers
If maxs || gs - gs*||< ε then Stop = true
End While
Attribute each observation Ri to the CLs class with
the condition Ri  CLs if μis=maxμir, 1≤r≤C
Figure 3: FCM algorithm
Despite the simplicity and the popularity of this algorithm, it suffers from the same drawbacks as the C means
does. These are: the number of classes must be known a priori, the initialization problem, the initial values taken by
the different variables, may have an impact on the convergence point, and the possibility that the convergence point
may stack on a local rather than a global optimum [6,7,8,9]. We present in the coming section an evolutionary
adaptation of this FCM based on an evolution strategy in order to get around these three drawbacks [10].
5. Evolutionary fuzzy classification
A classification method based on the evolutionary fuzzy C means has been reported in [10,18]. The method
looks for the optimal value for the centers of the classes in an iterative way. The algorithm adapted the population
elements constituted by class centers (gs)1≤s≤C , in such a way to minimize a fitness function defined on the search
space by [10,18]:
M C df
   is 
2
J ( chr )  Ri  g s
i 1 s 1
Where
1
 R  g 2 1 df
is   i s

C 1

k 1
 R  g
 i k
2  1 df


chr= (gsj)1≤s≤C,1≤j≤N
=( g11… g1N g21… g2N… gs1…. gSN… gC1… gCN)
Chr is a real line vector of size C×N. The genes are the (gsj)1≤j≤N components of the (gs) center.
At each iteration, a new population of centers is generated; it is formed by the best centers of the previous
generation. New center values are then produced by using the selection and the mutation genetic operators. The new
chromosome is obtained from the previous one by [10]:
gs* = gs + fm (gs0 - gs) × N (0,1)
Where fm is a weighting factor generally taken between 0.5 and 1, gs0 is the gravity center of the CLs class,
where CLs={Ri / μis=maxμir, 1≤r≤C}, CLs is formed by the Ri ’s for which the μis is the highest with respect to the
other classes. μis is the membership degree of the element represented by Ri to the class CLs. The strategic parameter
σ’= fm (gs0 - gs) is small when gs gets near to gs0, σ’ is high when gs is far away from gs0.
37
6. Classes number optimization

6.1. Xie and Beni criterion
Many criteria for determining the optimal number of classes have already been established [5,6]. Xie and Beni
criterion is based on a compacity and separability measure of the fuzzy partition. The compacity and separability
criteria that we used are defined by [3, 5]:
M C df
1
 is 
2
Comp (C )  Ri  g s
M i 1 s 1
And
2
Sep (C )  min s  s ' g s  g s '
Xie and Beni proposed the selection of Copt based on Copt  arg min C FXB (C )
Where
Comp(C )
FXB (C ) 
Sep(C )
6.2. Classes number optimization and classes’ respective center determination by evolution
strategy.
The EFCM algorithm is executed for several values of C, C varies between Cmin and Cmax and is incremented by
one at each step. For each value of C considered, we find the optimal centers and we compute the value of FXB. The
optimal number of classes is that for which FXB is minimal.
Once Copt is obtained, each object has already been attributed to its specific class. The attribution process is
made according to the value of the membership degree parameter. The object is attributed to the class CLS if the
membership degree of the object to this class is the highest.
R i   CL s
1 l ils
s 1
Where: l0=0, lCopt =M and l1, l2,…, lCopt-1 are integers that limits the Copt classes.
7. Selection of the most pertinent attributes for the classification
Our goal is to extract from N parameters (aj)1≤j≤N of an initial attribute vector Vinit, the q parameters, where q<<
N, which give the best discrimination of the different representative classes of the objects in the attribute space.
7.1. Total variance, intra-class and inter-class matrices
Let consider a set of M texture images {I1, I2, ..., Ii , ..., IM} characterized by N texture parameters written in a
line vector form Vinit = (a1 a2 ... aj ... aN). Let Ri = (aij) 1jN be a line vector of RN where aij is the value of the texture
parameter aj for the texture image Ii. Let EV = {a1, ..., aj, ..., aN}. Vinit is the attribute vector, Ri is the observation
associated with the image Ii or the realization of the attribute vector Vinit for this image, RN is the observation or
parameter space, EV is the set associated with Vinit. Each Ri belongs to the class CLs, s=1, …, C. Let r(aj ,al ) be the
linear correlation coefficient between two parameters aj and al of EV. Let V(m) be the vector formed by the mean value
of each parameter. Let V(m, CLs) be the vector formed by the mean value of each parameter of the CLs class
realizations. The total variance matrix TV associated with vector Vinit is defined by [1, 11]:
TV  1 Rit V(m) Rit V(m) 

M
t
M i 1
The intra-class matrix WV associated with vector Vinit is defined by [1,11]:
WV  1 
M s 1
C
 i
Rit V Rit V 
(m,CLs ) (m,CLs )
t
Ri CLs
38
WV is a compacity matrix [1,11]. The inter-class matrix BV associated with Vinit is defined by [1] [19] BV = TV - WV.
BV is a separability matrix [1,11]. The dimension of TV, WV and BV is NN, these matrices are used for establishing a
criterion to measure the discriminate power of a parameters space.
7.2. AURAY criterion
Let Vk (k N*) be an arbitrary attribute vector, of dimension q, extracted from Vinit. let TVk, WVk and BVk be
respectively the total variance, the intraclass and the interclass matrices associated with Vk, these matrices are all of
dimension q×q. the discriminate degree of Vk is estimated by [17]:
q
J Aur (q,Vk )  max  (T 1B )
l Vk Vk
l 1
Where λl (X) is the lth eigen value of the matrix X
This criterion is to be maximized.

7.3. Proposed criterion
We have improved the AURAY criterion by introducing the correlation as an additional factor for selecting the
parameters. This factor made it possible to avoid the information redundancy. Indeed the correlated parameters
provide similar information, considering all the correlated parameters in the selection procedure means an increase
of the parameters space dimension and an increase in the classifier complexity. This implies an increase in the
computation time, a slow decision process and a less precise selection of classes. If the dimension q space of the
selected attribute space contains k correlated parameters between them two by two, then k-1 parameters from among
them do not improve significantly the discrimination. In the same time the criterion value may increase by the
addition of these parameters. By removing these correlated parameters it would be possible to obtain a more
discriminate space with a lower dimension.
We have thus proposed the following criterion J:

q
max  (TV1 B )
l k Vk
l 1
J (q ,V ) 
k  r (a j , a )
l
(a , a )EV2
j l k
j l
This criterion J is to be maximized: an attribute vector Vk, extracted from Vinit, of dimension q (q<<N) is
optimal in the sense of the classification performances if its components maximize the inter-class distances (i.e.
move apart the class centers to the maximum distance), minimize the intra-class distances (i.e. get the points of each
class very close to the center) and are less correlated between them.
7.4. Optimization of the attribute vector by a genetic algorithm.

7.4.1. Proposed coding
Let Vk (k N*) be an arbitrary vector extracted from Vinit, Vk may have one to N components. To each Vk is
associated a chromosome chrk formed by N genes gkj :
chrk  ( g kj )1 j  N
Where
1 if a  E
g   j V
kj  k
 0 otherwise
chrk is a binary line matrix of N dimension. Each gkj gene codes the presence or absence of the aj parameter in
the Vk attribute vector. Vk coded by chrk may not be a possible solution unless if it satisfies the condition:
39
N
Dim(Vk )   g kj  q .
j 1
Any chromosome that does not satisfy this condition is discarded.
7.4.2. Proposed fitness function
For computing the selective value of chrk, we use the fitness function F defined by:
 r (a j , a )
l
(a j , a )  E 2
l V
k
jl
F ( chr ) 
k q
max  (TV 1BV )
l k k
l 1
Vk is the optimal solution if F(chrk) is minimal.

8. Experimental results
8.1. Texture images
8.1.1. Test1
We used forty five texture images for this test. For every value of C belonging to the set {2,3,4,5,6} the
classification of the test images is carried out by means of the EFCM algorithm, the data are classified in the Vinit
attribute space and the Xie and Beni criterion value, FXB, is computed for every value of C considered. Figure 4
shows the evolution of FXB with respect to C. the figure shows that the optimal number of classes for this case is
Copt=2.
Figure 4 : Evolution of FXB with respect to C, Copt=2.

The repartition of the images in their respective classes is: ( R i ) 1 i  30  CL 1 and ( R i ) 31  i  45  CL 2
Figure 5 shows two samples of the texture images used for different classes.
Wood texture  CL1 Liege texture  CL2

Figure 5: Two samples of texture images used
The genetic algorithm (GA) is used to extract the most pertinent two attribute parameters from among the
initially eleven ones. The GA reaches convergence rapidly; figure 6 shows the evolution of the selective value
associated with the best chromosome of the current population with respect to the generation number. The optimal
attribute vector Vopt obtained by GA is Vopt=(Uni Hom)
40
The results of the image classification in the (Uni Hom) space are show in fiqure 7. The figure shows that the
classes are well separated and the error rate of misclassifying the data is zero.
Figure 6: Fitness evolution for the J criterion.
Figure 7: Classes obtained in the (Uni Hom) space.
8.1.2. Test2
We used 105 texture images for this test. We considered the value of C to be 3,4,5 and 6. The images
classification was carried out by the EFCM algorithm and the images were represented in the Vinit attribute vector
space. FXB is computed for each value of C considered. Figure 8 shows the plot of FXB with respect to C. the figure
shows that the optimal number of classes for this case is Copt=4.

The repartition of the images in their respective classes is:
( Ri )1 i 15  CL1 , ( Ri )16  i  45  CL 2 , ( Ri ) 46  i  75  CL3 and ( Ri ) 76  i 105  CL4
Figure 9 shows four samples of the texture images used for different classes.
41
Wood1  CL1 Wood2  CL2 Liege CL3 Soil CL4
Figure 9: Four samples of texture images used

The next stage is to apply the genetic algorithm (GA) for extracting the most pertinent two parameters taken
from the initial eleven attributes considered. The GA convergence is reached quickly; figure 10 shows the evolution
of the best chromosome selective value associated with the current population with respect to the number of
generations. The optimal attribute vector Vopt obtained by GA is Vopt=(Hom Ent).

Images classification in the (Hom Ent) space is shown in figure 11. The classification error made is zero and the
classes are well separated.
Figure 11: Classes obtained in the (Hom Ent) space.
8.2. Application to the Alphabetic letters

We have used 120 images containing some of the Arabic letters. We considered several values for C, 2,3,4,5
and 6. The images classification was carried out by the EFCM algorithm and the images were represented in the Vinit
attribute vector space. FXB is computed for each value of C considered. Figure 12 shows the evolution of FXB with
respect to the number of classes C. The figure shows the optimal number of classes to be Copt=3.
42

The repartition of the images in their respective classes is:
( Ri )1 i  40  CL1 , ( Ri ) 41 i  76  CL 2 and (Ri )77i120  CL3
Figure 13 shows three samples of the texture images used for different classes.
Letter1  CL1 Letter2  CL2 Letter3  CL3

Figure 13: Three samples of alphabetic letters used
GA was then later applied to extarct the most pertient two parameters out of the 11 initial ones. Convergence
of the GA is obtained rapidly. Figure 14 shows the evolution of the best chromosome selective value assocciated
with the current population with respect to the number of generations. The optimal attribute vector obtained is
Vopt=(PN, Gy).

The results of the arabic characters classification in the (PN, Gy) space are shown in figure 15. The results show
that the two selected parameters PN and Gy obtain a good separation between the two classes in the parameters
space. We noticed that 4 characters out of 120 are not correctly classified which means an error rate of 3.33%.
43
Figure 15: Classes obtained in the (PN, Gy) space.
9. Conclusion
Non supervised data classification requires the selection of the most pertinent parameters which obtain a good
separation between the classes. We have proposed, for resolving this problem, an algorithm based on the
evolutionary approach. The EFCM algorithm combined with the Xie and Beni criterion have shown their ability to
capture the correct number of classes required for separating the data. The GA obtained the optimal attribute vector
for data classification in a rapid way. The technique proposed in this work has been validated for all the case studies
considered and have shown to be very efficient to obtain good results.
This technique may well be used concretely to resolve a problem in the field of quality control by artificial
vision.
References
[1] Cocquerez, T. P. and Phillip, S.: Analyse d'images : filtrage et segmentation. Editions MASSON, Paris (1995).
[2] P. Jiang, F. Ren and N. Zheng. ‘’A new approach to data clustering using a computational visual attention
model’’. International Journal of Innovative Computing, Information and Control Volume 5, Number 12(A),
December 2009.
[3] K. Tasdemir and E. Merenyi. A new cluster validity index for prototype based clustering algorithms based on
inter- and intra-cluster density. In Proc. Int’l Joint Conference on Neural Networks, 2007 (IJCNN 2007),
Orlando, FL, Aug 12-17, 2007.
[4] Bezdek, J., C. ‘’Cluster validity with fuzzy sets’’. J. Cybernetics, Vol.3, pp. 58-73, 1974.
[5] Xie, X. and Beni, G. ‘’ A validity measure for fuzzy clustering’’. IEEE Trans. On Pattern Analysis and machine
Intell., Vol.13, N° 8, pp; 841-847, 1991.
[6] Bezdek, J., C. and Hathaway, R., J. ‘’Optimisation of fuzzy clustering criteria using genetic algorithms’’. In proc.
First IEEE Conf. Evolutionary Computation, Piscataway, NJ: IEEE Pres., Vol.2, pp. 589-594, 1994.
[7] Kuncheva, L., I., and bezdek, J., C. ‘’ selection of cluster prototypes from data by a genetic algorithm’’ EUFIT,
pp. 1683-1688, Aachen, Germany, September 1997.
[8] Sarkar, M. Et al. ‘’A clustering algorithm using an evolutionary-based approach’’. Pattern Reconition Letters,
1997.
[9] Hall, L., Q., Ozyurt, I., B. and Bezdek, J., C. ‘’Clustering with a genetically optimized approach’’. IEEE Trans.
On Evolutionary Computation, Vol.3, N°2, pp. 103-110, july 1999.
[10] M. Nasri, M. El Hitmy, H. Ouariachi and M. Barboucha. ‘’Optimization of a fuzzy classification by
evolutionary strategies’’. In proceedings of SPIE Conf., 6th international Conference on Quality Control by
Artificial Vision, Vol. 5132, pp. 220-230, USA, 2003. Repulished as an SME Technical paper by the Society of
manufacturing engineers (SME), Paper number MV03-233, ID TP03PUB135, Dearborn, Michigan, USA, pp. 1-
11, 24 June 2003.
[11] Vandenbroucke, N. : Segmentation d’images couleur par classification de pixels dans des espaces d’attributs
colorimétriques adaptés : Application à l’analyse d’images de football. Thèse de Doctorat, Université de Lille 1,
France, 14 Décembre (2000).
[12] Haralick,, R. M.: Statistical and structural approaches to texture. Proceedings of the IEEE,
vol. 67, N° 5, pp. 786-804 (1979).
[13] Galloway, M. M.: Texture analysis using gray level run lengths. Computer Graphics and Image Processing
(CGIP), Vol. 4, pp. 172-179 (1975).
44
[14] Postaire, J. G. : De l'image à la décision : analyse des images numériques et théorie de la décision. Editions
DUNOD (1987).
[15] Presberger, T. et Koch., M. ‘’ Comparison of evolutionary strategies and genetic algorithms for optimization of
a fuzzy Controller ’’. Proc. of EUFIT’95, Aachen, Germany, August 1995.
[16] Lutton, E. : Algorithmes génétiques et fractales. Dossier d'Habilitation à diriger des recherches, Université Paris
XI Orsay, 11 Février (1999).
[17]Auray, J. P., Duru, G. and Zighed, A.: Les méthodes d’explication. In : Analyse des Données
Multidimensionnelles. Alexandre Lacassagne, Lyon (1991).
[18] Nasri. M. ‘’ contribution à la classification de données par Approches Evolutionnistes : Simulation et
Application aux images de textures’’. Thèse de doctorat. Université Mohammed premier Oujda. 2004.
[19] Haitaamar, S. ‘’Segmentation de textes en caractères pour la reconnaissance optique de l’écriture arabe’’. Thèse
de doctorat, Université El-Hadj Lakhdhar Batna, 2007.
[20] Ghaouth Belkasmi. M. ‘’ Reconnaissance des formes par réseaux de neurones : application aux caractères
arabes’’. Thèse de DESA. Université Mohammed premier Oujda. 2007.
45
Segmentation of Textured Images using Orientation as Feature Vector for

Defect Detection in Wood and Textile Industries
Punamchand M. Mahajan 1, Dr. Satish R. Kolhe 2, Dr. Pradeep M. Patil 3
1
Department of Electronics and Telecommunication Engineering, J.T.Mahajan College of
Engineering, Faizpur, India E-mail: pmmahajan@yahoo.com
2
Department of Computer Science, North Maharashtra University, Jalgaon, India,
E-mail: srkolhe2000@gmail.com
3
Department of Electronics Engineering, Vishwakarma Institute of Technology, Pune, India, E-
mail: patil_pm@rediffmail.com
Abstract
Quality control activities form an essential information feedback loop for any business, with potential influence
on the design, process planning, logistics functions as well as manufacturing. Visual inspection is an important part
of quality control in the industries like textile and wood. Human inspection may not achieve high degree of accuracy.
In order to increase accuracy, attempts are being made to replace manual inspection by Automated Visual Inspection
(AVI), where the texture based segmentation plays a vital role. In almost all the applications of texture segmentation,
segmentation is achieved according to the human perception system i.e. all feature vectors are calculated using
standard algorithms such as Gabor filter, Wavelet Transform, Morphological filter etc. There is no necessity of
calculating all the feature vectors of an image in the applications like defect detection, only local disorder calculation
is sufficient. In the textured images where directionality is the main concern, it is very necessary to concentrate on
the dominant orientation of the regions. “Segmentation of textured images using orientation as feature vector for
defect detection in wood and textile industries” is an effort to concentrate on orientation as the main feature vector
which increases the computational speed, reliability as well as reduces the memory requirement hence feasible for
real-time factory implementations. Experimentations for proposed algorithm were carried out on synthetic images,
natural images and standard images from Brodatz album. The proposed algorithm is powerful for defect detection
and the parameters are set based on the application in hand. Step and window size has to be properly tuned while
calculating the orientation image. The number of clusters used for the classification plays a vital role in final
segmentation results.
Keywords: Directionality, Dominant orientation image, Defect detection, Feature vector, Clustering, classification,
and Texture segmentation.
46
1. Introduction
Image segmentation is the most important stepping stone leading to analysis of processed image data. Its main
goal is to divide an image into various parts that have a strong correlation with the objects / areas of real world
containts in that image. Many approaches have been used to perform image segmentation depending on the
information representation in images e.g. when edges are not directly visible but can be interpreted as the limit
between regions; a region-based segmentation is more efficient than an edge-based segmentation [1]. Textural
properties of regions are referred many times while considering image segmentation. Texture segmentation is an
important area in image processing. Basically, it aims at dividing a textured image into several regions with the same
texture features. Texture refers to properties that represent the surface or structure of an object. An image region has
a constant texture if a set of its local properties in that region is constant, slowly changing or approximately periodic
[2]. The set of local properties can be understood as primitive types and their spatial relationships. Segmentation of
textures requires identification of proper texture-specific features with good discriminative power. Three most
important dimensions of natural texture discrimination are periodicity, directionality and granularity [3]. In almost all
the applications of texture segmentation, segmentation is achieved according to the human perception system. All
feature vectors are calculated by using standard algorithms such as Gabor filter [4], Wavelet transform [5],
Morphological filter [6,7] etc.
Quality control plays an important role in any business. Visual inspection is an important part of quality control in
the industries like textile, wood, paper, steel, leather etc. This inspection has traditionally been performed using the
human eye. But besides the lack of reliable quantitative measurement data, visual assessment made by humans has
been time consuming and cost-intensive. Moreover, due to the stress, human inspection does not achieve a high
degree of accuracy and preciseness. To increase accuracy, attempts are being made to replace manual inspection by
Automated Visual Inspection (AVI) which employs a camera and imaging routines [4]. The process accompanying
image acquisition, automatic evaluation and control could form the basis for a system that will ensure a very high
degree of product quality control. The main difficulty is to solve the problem of quantifying visual impressions in
complex situations like those met in fabric manufacture.
Thus segmentation of textures is very important to decide the quality of product. Generally all the feature vectors
are considered for texture segmentation in order to classify the exact region from an image [4,5]. Due to this the
algorithm becomes time consuming and computationally complex also requires more memory. In purview of the
present industry demands, attempts are being made to reduce the complexity of the algorithms. There is no necessity
of calculating all the feature vectors of an image in the applications like defect detection, only the local disorder
calculation is sufficient. In proposed algorithm only orientation feature is used for calculating the local disorder
which is approximated by the dominant orientation in particular region for texture segmentation.
Understanding the effect of orientation on texture segmentation is essential due to its neurophysiologic basis, its
central role in perceptual organization, and its close relationship to shape perception [8]. The description and
modeling of textures is still a great challenge. Over the last decade the automatic detection of errors in textures has
gained increasing importance for industrial applications. Manufacturing high quality products needs quality control.
In many areas of production analyzing planar surfaces of the product under test can assess the quality. Many papers
develop special solutions for specific error detection tasks, e.g. the detection of flaws on woven textiles and roll
goods. Campbell and Murtagh [9] proposed an AVI system for defect detection in woven textiles. Kallmes,
Scharcanski and Dodson [10] proposed a method which can detect flaws in nonwoven textiles. Jain and Farrokhnia
[4] used texture features computed from a bank of Gabor filters to automatically classify the uniformity of painted
metallic surface. Jain, Ross and Prabhakar [11] have used a structural approach for fingerprint matching in textured
images. Hanbury and Serra [6] have utilized morphological based texture analysis methods to detect defects in wood
automatically. Dmitry Chetverikov [12] has done work on texture analysis using feature-based pair wise interaction
maps. Chetverikov and Hanbury [3] used regularity and local orientation as local properties to find defects in textile
images. Yeh and Perng [13] worked for defect compensation in leather industries. Lovergine, Branca, Attolico and
Distante [7] have performed oriented texture analysis with a morphological approach for leather inspection. Allan
Hanbury with Frederic Gardeux [14] worked in the field of automated wood inspection for the industry namely
Scanwood System in France and used the orientation as local property for feature extraction with the morphological
approach for segmentation. Patil and Sontakke proposed clustering and classification algorithms [15, 16]. To
increase the acceptance for automatic visual inspection systems in a wide industrial field it is necessary to develop
method that can be easily adapted. Developing only special solutions for every single error detection task produces
high costs which are unacceptable for the manufacturer. In this paper a method for segmentation of weakly oriented
47
textures where segmentation is scale independent is proposed.

Remaining part of the paper is organized as: Section II concentrates on the algorithm used to calculate feature
vector and clustering as a classifier. Section III deals with the simulation results and discussion. Conclusions along
with the idea for further work are listed in section IV. Finally, a few references along with an appendix containing
various natural and standard images, processed using proposed algorithm are attached.
2. Theoretical aspects
The important steps involved in proposed orientated texture image segmentation algorithm are filtering, extraction
of feature vectors and application of classifier as shown in Fig.1. The routines or ideas developed for each step are
void, but the fundamental steps remain the same.
Source image may contain noise. Hence after receiving the input data, the first step is to remove the unwanted
energy or signal. Filters are applied on the noisy image to enhance the image. Numbers of filters are available in the
field of image processing. Principle objective of enhancement is to preprocess an image so that the resultant image is
more suitable than the original image for further processing. Smoothing filters, usually low pass filters are useful in
texture segmentation for blurring the image and reducing the noise. Blurring removes the small details from an
image and bridges the small gaps in lines or curves. Gaussian low pass filter [9] is preferred because the shapes of
filters based on Gaussian function are easily specified and also the Forward and Inverse Fourier Transforms of a
Gaussian function are real Gaussian functions. The second important step in the segmentation is feature vector
extraction. A convenient representation of oriented texture is a two-dimensional vector field. Each vector is
associated with a point in the texture: the direction of vector represents the orientation at that point and its length is a
measure of the orientation strength or coherence. The vector field can be encoded by two grayscale images; an
orientation image [14], in which the direction of each vector is represented by the pixel values, and a corresponding
magnitude with the vector strength image [14]. Several algorithms exist for calculating the vector fields of the
oriented textures. In this paper Gradient based vector fields are used for calculating orientation image and strength
image. Using these images the dominant oriented image is calculated in two sub steps as:
Step 1: For each pixel k, l  , horizontal and vertical gradients H kl and Vkl are calculated from the preprocessed
image as,
H kl  F (k , l )  F (k , l  1) (1)
Vkl  F (k , l )  F (k  1, l ) (2)
From these gradients the modulus Rkl and the angle  kl (between 00 and 3600) are calculated using
2 2
Rkl  H kl  Vkl (3)
Vkl
 kl  tan 1 (4)
H kl
For computational efficiency the gradient amplitude can be approximated by the magnitude combination as,
Rkl  H kl  Vkl (5)
Step 2. Once the angle values at each pixel location is calculated, the next sub step is to calculate the dominant
orientation image. A neighborhood of width  2 h and height  2 v is moved over the image in steps of size  h
pixels horizontally and  v pixels vertically. At each window position ( x, y ) , where x and y are the co-ordinates

of pixels of dominant orientation image and are multiples of  (if  h   v ) . The dominant local orientation O xy
(between 00and 1800) is calculated using
 1 
 arctan 
  k ,l W

Rkl sin 2 kl 
2

O xy
2 

 R
 k ,l W kl
2
cos 2 
kl


(6)
An orientation of 900 indicates vertical structures and orientation of 00 & 1800 indicate horizontal structures. The
veins are undirected lines which indicate the 1800 periodicity of the calculated orientations [6].
The final step in the segmentation algorithm is the use of classifier. In this paper k-means clustering algorithm [2]
is used as classifier for finding data clouds in a given data set as shown in Fig. 2. Basically this clustering algorithm
48
is used to distribute the N feature vectors into N-dimensional feature vector space. As only one feature vector
(orientation) is present; distribute this feature vector into k clusters in single dimensional feature space. Label each
feature vector belonging to a definite cluster according to the minimum distance classification algorithm [17]. The
Euclidean distance of each feature vector from the k cluster-centers is computed and the vector is assigned to the
cluster having its centre at a minimum distance. The k-means clustering algorithm is thus an iterative procedure of
vector labeling, till a stable cluster assignment is achieved. Generally, k initial centers are computed as k
arithmetic means between the vector components. To start with, a cluster assignment is made based on the initial
cluster centers. The cluster centers are updated from the newly obtained cluster assignments by computing the
centroids of the cluster members. The vectors are again labeled according to the updated cluster centers. This
procedure of cluster centre updating and vector labeling is repeated till the cluster assignment becomes a stable
structure and does not change furthermore. At the end of the whole algorithm the labeled image is converted into rgb
image so that the segmented image becomes visually clear.
3. Results and Discussion

The proposed algorithm was executed on 2.4GHz, 256MB RAM Pentium-IV system with Windows XP platform.
The experimentations were carried out on three types of databases: Synthetic database or database of test images,
natural images and images from Brodatz database [18]. A synthetic database having images S1 to S 25 was created
where S stands for synthetic. Microsoft paint editor was used for creating the images. These images contain lines
with different orientations, may be scaled or may contain some curvature lines along with different grayscale
intensities were normalized to the intensity range of [0, 255]. The second database used was of some natural images
e.g. the wooden board images scanned in laboratory. The third database used was Brodatz album which is considered
as standard database in image processing and pattern recognition fields. Orientation images from Brodatz album
were used. Four patterns of different orientations from the album were cropped in order to construct 256×256 image
for the experimentation purpose. The single image contains two or more classes of orientation patterns. These types
of textures are characterized by a high level of orientation specifically at each point of texture.
Fig. 3(a) shows the synthetic image of size 256  256. This image was smoothened using Gaussian filter by
selecting the values of standard deviation   1.4 and the window size of 5 as shown in Fig. 3(b). It can be
observed clearly that the small gaps between the lines were filled. This pre-processed image was used to calculate the
horizontal and vertical gradients using (1) and (2) as shown in Fig. 3 (c) and (d) respectively. Use (3) and (4) to
calculate magnitude Rkl and phase  kl values at each pixel. Using these values dominant local orientation was
calculated by (6). A neighborhood of width  2 h = 16 and height  2 v = 16 was selected and moved over the image in
steps of size  h = 4 pixels horizontally and  v = 4 pixels vertically. At each position ( x, y ) of neighborhood, the
 
dominant orientation O xy (0  O xy  180) was calculated. After calculation of orientation image a reduced size
image of 56  56 pixels was obtained as shown in Fig. 3(e). k -means algorithm was used for segmentation by
selecting the number of clusters equal to 4. The clustered image is then converted into rgb image as shown in Fig.
3(f). From Fig. 3(a) and (f) it can be judged that the implemented scheme is the powerful algorithm.
In Fig. 4(a) and (b), some patches are observed at the boundary when there is abrupt change in orientation. This is
due to the overlapping of mask moved over the whole image. Even though there is abrupt change in orientation at
boundary, the mask considers the angles from both sides and interprets it as a different angle. By selecting proper
size of mask patches can be reduced.
The number of clusters k used in the classifier affects the final segmentation results as shown in Fig. 5. If the
number of clusters are less, classification of regions is improper, leads to under-clustering whereas if number of
clusters are more over-clustering is observed. The selection of k depends on the input image complexity. The same
algorithm was applied for the natural images as well as images from Brodatz album. The results are shown in
Appendix.
49
3.1 Figures and Tables
Input Image (Source Image)
Filtering
Feature Vector Formation
Classifier
(Fuzzy k means clustering)
Segmented Image
Figure 1: Flow-chart for segmenting the textured image.
Figure 2: Flowchart of K-Means Clustering Algorithm
50
(a) (b) (c)
(d) (e) (f)
Figure 3: Images explaining each step of the proposed algorithm
patch
Figure 4: Patch due to improper selection of mask size
(a) (b)
(c) (d)
Figure 5: Effect of K on segmentation results
(a) Synthetic image (b) Under-clustering
(c) Proper clustering (d) Over-clustering
4. Conclusion
The fundamental structural feature i.e. orientation can provide a framework for reasonably general approach to detect
structural variations. The algorithm has a number of parameters which can be adjusted depending on the application
in hand. Standard deviation shows scale invariance to the orientation in textured image. The step and mask sizes
51
affects performance of the algorithm. To get orientation image of required size, the step size has to be adjusted
properly. To cover all the pixels from the original image, the step size should be less than or equal to the size of
mask. Step size and window size of the mask can be tuned in such a way that size of mask should be large enough to
cover the defect at least more than two times in horizontal as well as vertical scanning for selected step size and mask
size should be small enough to reduce the width of patch at abrupt changes in orientations in order to detect exact
region boundaries. The number of clusters used for the classification plays a vital role in final segmentation results.
Considering the results obtained with respect to computational complexity, memory requirement and reliability, the
orientation approach seems to be feasible for real-time factory implementations.
The orientation approach is more local and appropriate for the unidirectional flow like textures. Further
refinements in the results can be achieved by using the regularity as an additional feature vector to extend the use of
algorithm for regular textured images is the subject of future work.
APPENDIX
52
5. References
[1] Gonzalez R. and Woods R., (1996), Digital Image Processing, Third Edition, Addison Wesley.
[2] Milan Sonka, Vaclav Hlavac and Rojer Boyle, (1999)Image Processing, Analysis, and Machine Vision, Second
Edition, PWS Publishing..
[3] D. Chetverikov and A. Hanbury, (2002), “Finding defects in texture using regularity and local orientation”,
Pattern Recognition, 35(10).
[4] Jain, A.K. and F. Farrokhnia,(1991), “Unsupervised Texture Segmentation Using Gabor Filters,” Pattern
Recognition, 24(2), 1167-1186.
[5] A. Laine and J Fan, (1993), “Texture classification by Wavelet Packet Signatures”, IEEE Trans. On PAMI, 15,
1186-1191.
[6] A.G. Hanbury and J. Serra, (2002), “Analysis of Oriented Textures using Mathematical Morphology”, Vision
with Non-Traditional Sensors, Graz, Austria.
[7] F.P. Lovergine, A. Branca, G. Attolico, A. Distante, (1997), “Leather inspection by oriented texture analysis
with a morphological approach”, International Conf. on Image Processing.
[8] P. Kruizinga and N. Petkov, (1998), “Grating cell operator features for oriented texture segmentation”, Proc. of
14th Int. Conf. on Pattern Recognition, Australia, August 16-20, 1010-1014.
[9] J. G. Campbell and F. Murtagh, (1998), “Automatic visual inspection of woven textiles using 2-D stage defect
detector”, Optical Engineering, 37(9), 2536-2542.
[10] A.Kallmes, J. Scharcanski and C.T.J. Dodson, (2000), “Uniformity and Anisotropy in Nonwoven Fibrous
Materials”, International Nonwovens Technical Conference, Dallas.
[11] Anil Jain, Arun Ross and Salil Prabhakar, (2001),“Fingerprint Matching using Minutiae and Texture Features”,
Proc. of Int’l Conference on Image Processing, Thessaloniki, Greece, Oct 7-10, 282-285.
[12] Dmitry Chetverikov, (1999), “Texture analysis using feature-based pair wise interaction maps” Pattern
Recognition, 32, 1999, 487-502.
[13] C. Yeh, D.B. Perng, (2000), “A reference standard of defect compensation for leather transactions”, The
International Journal of Advanced Manufacturing Technology, Springer-Verlag London Ltd.
[14] Allan Hanbury and Frederic Gardeux, (1998), “A Quantative Description of Wood Textures”, Work done at
Scanwood Systems, Pont-a-Mousson, France.
[15] P.M. Patil, P.S. Dhabe, U.V. Kulkarni and T.R. Sontakke, (2003), “A New Clustering Method Based on
Geometrical Moment of Patterns”, In proceedings of the IEEE 12th International Conference on Fuzzy Systems,
FUZZ-IEEE 03, Missouri, USA., 2, 887-891.
[16] Pradeep M. Patil, and T.R. Sontakke, (2007), “Rotation Scale and Translation Invariant Handwritten Devanagiri
Numeral Character Recognition using General Fuzzy Neural Network,” Pattern Recognition Journal (Elsevier
Science), Volume 40, Issue 7 , pp 2110-2117.
[17] Jain A.K., (1997), Fundamentals of Digital Image Processing, Prentice Hall Of India, 1997.
[18] P. Brodatz, (1966), “Textures: A photographic album for artists and designers”, Dover Publications, Inc., New
York.
53
Fuzzy Time Series Forecasting with better accuracy from Mean with Varied
Intervals
*Arutchelvan.G, **Dr. Srivatsa.S.K and ***Dr. Jagannathan.R
* Associate Professor and HOD, Department of Computer Science and Applications,

Adhiparasakthi College of Arts and Science, G.B.Nagar,Kalavai-632506,Tamilnadu ,
E-Mail: garutchelvan@yahoo.com
** Professor(Retired), Department of Electronics and Engineering,
Madras Institute of Technology, Anna University, Chennai ,India, Tamilnadu ,
E-Mail :profsks@rediffmail.com
*** Deputy Registrar (Education), Vinayaka Missions University, Chennai -10
E-Mail: jaganr1@yahoo.com
Abstract
Forecasting task is an important method in our daily life. Various methods are available for forecasting the value
from the existing data. Forecasting value by using fuzzy time series is one of the method based on partitioning the
data into different interval. In this paper, we predict the new value which is very close to the original data, based on
partitioning the input data into different interval. Our method used the historical data of birth rate of Tamil Nadu.
The proposed method belongs to the kth order and time-variant methods. The proposed method gives the best
forecasting accuracy rate.
Key words — Fuzzy sets, Fuzzy groups, fuzzified data, fuzzy time series.
54
1. INTRODUCTION
Last two decades, Various approaches have been used to forecasting value using time series. Among them, ARMA
models and Box-Jenkins model are famous approaches. But, the classical time series methods can not deal with
forecasting problems in which the values of time series are linguistic terms represented by fuzzy sets [11], [23].
Therefore, Song and Chissom [18] proposed the new approach of theory of fuzzy time series to overcome this
drawback. Based on this, Song et al. proposed few forecasting methods [16], [18], [19], [20] to forecast the
enrollments of the University of Alabama. In [1] Chen and Hsu and in [2], Chen attempted a method to forecast the
enrollments of the University of Alabama using fuzzy times series. Main advantage of this method reduces the
calculation, time and minimizing the calculation process. In [8], Hwang, Chen and Lee proposed a new methods
using differences of the enrollment to forecast the enrollments data of the University of Alabama based on fuzzy
time series. In [5] and [6], Huang adopted heuristic rules to forecast the enrollments using [2]. In [4], Chen used a
forecasting methods based on high-order fuzzy time series for forecasting the enrollments of the University of
Alabama. In [3], Chen and Hwang proposed a method based on fuzzy time series to forecast the daily temperature.
In [15], Melike and Konstsntin gave a new first order time series model for forecasting enrollments of the University
of Alabama. In [14], Li and Kozma proposed a dynamic neural network method for prediction using the KIII
model. In [21], Su and Li proposed a new method for fusing global and local information in predicting time series
based on neural networks. In [22], Sullivan and Woodall remodelled the first-order time-variant fuzzy time series
model. In[18], song and chisson remodelled the first-order time-invariant fuzzy time series. In [13], Lee, Wang and
Chen proposed two factor high order fuzzy time series for forecasting daily temperature in Taipei and TAIFEX. In
[9], Jilani and Burney and in [10], Jilani, Burney and Ardil presented new fuzzy metric approach for high order
multivariate fuzzy time series forecasting for car road accident in Belgium.
In this paper, we present a new approach for prediction the value of birth rate of Tamil Nadu and then compared
to original value. Our proposed method is k step first-order univariate time-variant method. The proposed method
gives best forecasting accuracy rate when compared with different intervals. The rest of this paper is organized as
follows. In section 2, we briefly describe the basic concepts of fuzzy time series. In Section 3, we present our
method of fuzzy forecasting based on means of partitioning the input data with different intervals. In Section 4, we
compared the forecasting results of the proposed method with the existing original data. In section 5, we conclude
the results.
2. BASIC CONCEPTS OF FUZZY TIME SERIES:

There are number of definitions for fuzzy time series.
Definition: Let Y(t), (t=...0,1,2,…) be the universe of discourse and Y(t) is subset of R . Assume that fi(t), i=1,2,…
is defined in the universe of discourse Y(t) and F(t) is a collection of f(ti), ( i=...0,1,2,…), then F(t) is called a fuzzy
time series of Y(t), i=1,2,... Using fuzzy relation, we define F(t)=F(t-1) o R(t,t-1) where R(t,t-1) is a fuzzy relation
and “o” is the max-min composition operator, then F(t) is caused by F(t-1) where F(t) and F(t-1) are fuzzy sets. Let
F(t) be a fuzzy time series and let R(t, t-1) be a first order model of F(t). If R(t, t-1)=R(t-1,t-2) for any time t then
F(t) is called a time-invariant fuzzy time series. If R(t,t-1) is dependent on time t, that is, R(t, t-1) may be different
from R(t-1,t-2) for any t, then F(t) is called a time-variant fuzzy time series.
55
3. PROPOSED METHOD USING FUZZY TIME SERIES:

In this section, we present our method to forecast the data of birth rate based on means. The historical data of birth rate
are given in Table-1. First, based on [1], we defined the partition the universe of discourse into equal length intervals. Then
based on means of original data and means of frequency of interval data, we redefine the intervals. After this, define some
membership function for each interval of the historical data to obtain fuzzy data to form a fuzzy time series. Then, it
establishes fuzzy logical relationships (FLRs) based on the fuzzified data in Table-4. Finally, it uses our proposed method.
Step 1: Define the universe of discourse U and partition it into four equal intervals u1,u2,u3,……un. For example, assume
that the universe of discourse U=[15, 35] is partitioned into four even and equal length intervals.
Step 2: Get a mean of the original data. Get the means of frequency of each interval. Compare the means of original and
frequency of each interval and then split the four interval into number of sub-interval respectively.
Step 3: Define each fuzzy set Ai based on the re-divided intervals and fuzzify the historical data shown in Table-1, where
fuzzy set Ai denotes a linguistic value of the birth data represented by a fuzzy set. We have used triangular membership
function to define the fuzzy sets Ai [10]. The reason for fuzzifying the birth data into fuzzified data is to translate crisp
values into fuzzy sets to get a fuzzy time series.
Step 4: Establish fuzzy logical relationships based on the fuzzified data where the fuzzy logical relationship “Ap, Aq, Ar ->
As “ denotes that “if the fuzzified data of year p, q and r are Ap, Aq and Ar respectively, then the fuzzified data of the year (r)
is Ar”. where aj-1,aj, aj+1 are the mid points of the fuzzy intervals Aj-1, Aj, Aj+1 respectively. Based on the fuzzify birth data
obtained in step 3, we can get the fuzzy logical relationship group (FLGR) as shown in Table-4.
Divide each interval derived in step 2 into subintervals of equal length with respect to the corresponding means of the
interval. We have assumed thirty five partitions of the universe of discourse of the main factor fuzzy time series. Assuming
that 0 ≠ Ai, for every Ai, i=1,2,…..35.
1+0.5
---------------------- if j=1,
(1/ a1 ) +(0.5/ a2 )
0.5 + 1 + 0.5
--------------------------------------- if 2 <= j <= n-2
tj = (0.5/ aj-1 ) +( 1/ aj ) + (0.5/ aj+1 )
0.5 + 1
----------------------- if j= n
(0.5/an-1)+ (1/ an )
TABLE - 1
YEARLY BIRTH RATE FROM 1971 TO 2006 IN TAMIL NADU.
--------------------------------------------------------------------- ---------------------------------------------------------------------
YEAR Birth Rate
1971 31.4
1972 32.4
1973 30.0
1974 30.7
1975 30.7
1976 30.7
1977 29.8
1978 28.8
1979 28.9
1980 27.9
1981 28.0
56
1982 27.7
1983 27.7
1984 28.0
1985 24.7
1986 23.8
1987 24.0
1988 22.7
1989 23.1
1990 21.6
1991 20.8
1992 20.7
1993 19.5
1994 19.2
1995 20.2
1996 19.5
1997 19.0
1998 19.2
1999 19.3
2000 19.3
2001 19.1
2002 18.5
2003 18.3
2004 17.1
2005 16.5
2006 16.2
TABLE – 2
MEANS OF ORIGINAL DATA AND FREQUENCY OF INTERVALS DATA
Interval Re-divide the interval

[15,20] 10
[20,25] 8
[25,30] 9
[30,35] 8
TABLE - 3
FUZZY INTERVALS USING MEAN BASED PARTITIONING
Linguistics Intervals
u1 [15, 15.5]
u2 [15.5, 16.0]
u3 [16.0, 16.5]
u4 [16.5, 17.0]
u5 [17.0, 17.5]
u6 [17.5, 18.0]
u7 [18.0, 18.5]
u8 [18.5, 19.0]
u9 [19.0, 19.5]
u 10 [19.5, 20.0]
u 11 [20.0, 20.625]
u 12 [20.625, 21.25]
u 13 [21.25, 21.875]
u 14 [21.875, 22.5]
57
u 15 [22.5, 23.125]
u 16 [23.125, 23.75]
u 17 [23.75, 24.375]
u 18 [24.375, 25.0]
u 19 [25.0, 25.56]
u 20 [25.56, 26.12]
u 21 [26.12, 26.68]
u 22 [26.68, 27.24]
u 23 [27.24, 27.8]
u 24 [27.8, 28.36]
u 25 [28.36, 28.92]
u 26 [28.92, 29.48]
u 27 [29.48, 30.0]
u 28 [30.0, 30.625]
u 29 [30.625, 31.25]
u 30 [31.25, 31.875]
u 31 [31.875, 32.5]
u 32 [32.5, 33.125]
u 33 [33.125, 33.75]
u 34 [33.75, 34.375]
u 35 [34.375, 35.0]
4. A COMPARISON OF DIFFERENT FORECASTING METHODS
In the following Table-5 summarizes the forecasting results of the proposed method from 1971 to 2006, where the
universe of discourse is divided into thirty five intervals based on means based partitioning. In the following, we use the
average forecasting error rate (AFER) and mean square error (MSE) to compare the forecasting results of different intervals,
where Ai denotes the actual value and Fi denotes the forecasting value of year (i), respectively.
| Ai – Fi | / Ai
AFER = ------------------ * 100 %
n
n
∑ ( Ai – Fi ) 2
i=1
MSE = --------------------
n
In Table -6, we compare the forecasting results of the proposed method of different intervals with original data . From
Table -3, we can see that when the number of intervals in the universe of discourse is thirty five intervals are sub partitioned
based on mean. By using C program to calculate the values of the MSE and AFER of different intervals . That is, the
proposed method can get a higher accuracy rate for forecasting birth rate of Tamilnadu.
TABLE -4
THIRD ORDER FUZZY LOGICAL RELATIONSHIP GROUPS
Group 1: X3,X5,X7 -> X7
Group 2: X5,X7,X7 -> X9
Group 3: X7,X7,X9 -> X9
Group 4: X7,X9,X9 -> X9
Group 5: X9,X9,X9 -> X9
58
Group 6: X9,X9,X9 -> X8
Group 7: X9,X9,X8 -> X9
Group 8: X9,X8,X9 -> X11
Group 9: X8,X9,X11 -> X9
Group 10: X9,X11,X9 -> X9
Group 11: X11,X9,X9 -> X12
Group 12: X9,X9,X12 -> X12
Group 13: X9,X12,X12 -> X13
Group 14: X12,X12,X13 -> X15
Group 15: X12,X13,X15 -> X15
Group 16: X13,X15,X15 -> X17
Group 17: X15,X15,X17 -> X17
Group 18: X15,X17,X17 -> X18
Group 19: X17,x17,x18, -> X24
Group 20: X17,X18,X24 -> X23
Group 21: X18,X24,X23 -> X23
Group 22: X24,X23,X23 -> X24
Group 23: X23,X23,X24 -> X24
Group 24: X23,X24,X24 -> X25
Group 25: x24,X24,X25 -> X25
Group 26: X24,X25,X25 -> X27
Group 27: X25,X25,X27 -> X29
Group 28: X25,X27,X29 -> X29
Group 29: X27,X29,X29 -> X29
Group 30: X29,X29,X29 -> X27
Group 31: X29,X29,X27 -> X31
Group 32: X29,X27,X31 -> X30
59
TABLE -5
Actual Mid Calculated | Ai -Fi|
Years Values Values of Values (Ai-Fi) 2 -------
Ai Interval Fi Ai
1971 31.4 31.5625 31.55631 0.024433 0.004978047
1972 32.4 32.1875 32.18143 0.047772 0.00674596
1973 30.0 29.74 29.74293 0.066087 0.008569148
1974 30.7 30.9375 30.93119 0.053447 0.007530475
1975 30.7 30.9375 30.93119 0.053447 0.007530475
1976 30.7 30.9375 30.93119 0.053447 0.007530475
1977 29.8 29.74 29.74293 0.003257 0.001915249
1978 28.8 28.64 28.63452 0.027382 0.005745691
1979 28.9 28.64 28.63452 0.070477 0.009186018
1980 27.9 28.08 28.07441 0.030421 0.006251428
1981 28.0 28.08 28.07441 0.005538 0.002657673
1982 27.7 27.52 27.5143 0.034484 0.00670393
1983 27.7 27.52 27.5143 0.034484 0.00670393
1984 28.8 28.08 28.07441 0.005538 0.002657673
1985 24.7 24.6875 24.67185 0.000792 0.001139482
1986 23.8 24.0625 24.05438 0.064709 0.010688251
1987 24 24.0625 24.05438 0.002957 0.002265849
1988 22.7 22.8125 22.80394 0.010803 0.004578641
1989 23.1 22.8125 22.80394 0.087654 0.012816661
1990 21.6 21.5625 21.55344 0.002168 0.002155638
1991 20.8 20.9375 20.92817 0.016427 0.006161898
1992 20.7 20.9375 20.92817 0.05206 0.011022584
1993 19.5 19.25 19.2435 0.0579 0.013153626
1994 19.2 19.25 19.2435 0.001893 0.002265849
1995 20.2 20.3125 20.31946 0.01427 0.005913765
1996 19.5 19.25 19.2435 0.06579 0.013153626
1997 19.0 18.75 18.74333 0.065879 0.013508897
1998 19.2 19.25 19.2435 0.001893 0.002265849
1999 19.3 19.25 19.2435 0.003192 0.002927238
2000 19.3 19.25 19.2435 0.003192 0.002927238
2001 19.1 19.25 19.2435 0.020593 0.007513314
2002 18.5 18.25 18.24315 0.065973 0.013883886
2003 18.3 18.25 18.24315 0.003232 0.00310666
2004 17.1 17.25 17.24275 0.020378 0.008347987
2005 16.5 16.25 16.2423 0.066407 0.001561793
2006 16.2 16.25 16.2423 0.00179 0.002611361
MSE = 0.03189 AFER = 0.674256686 %

TABLE -6
Intervals Mean Square Forecasting Error
10,7,4,4 0.074596 0.908223
10, 8, 6, 5 0.032691 0.645987
10, 8, 9, 8 0.03189 0.674256686
10,10,10,10 0.021124 0.532233
10,12,14,10 0.016730 0.464739
10,14,16,11 0.014882 0.440870
60
5. Conclusions
In this paper, we have presented mean based partitioning of the birth rate data of Tamil Nadu. The proposed methods
belong to the first order and time-variant methods. From Table-6, we can see that the AFER and MSE of the forecasting
results with different intervals. The correct forecasting values are found with smallest values of AFER and MSE. We have
got correct forecasting values with the intervals 10, 14, 16, 11.
REFERENCES
[1] Chen S.M. and Hsu C.-C. 2004. A new method to forecasting enrollments using fuzzy time series, International Journal
of Applied Science and Engineering, 2,3 : 234-244.
[2] Chen, S.M. 1996, Forecasting enrollments based on fuzzy time series. Fuzzy Sets and Systems, 81: 311-319.
[3] S.M. Chen and J.R. Hwang, “Temperature prediction using fuzzy time series”, IEEE Transactions on Systems, Man, and
Cybernetics- Part B: Cybernetics, Vol.30, pp.263-275,2000.
[4] S.M. Chen, “Forecasting enrollments based on high-order fuzzy time series”, Cybernetics and Systems: An International
Journal, Vol. 33: pp 1-16, 2002.
[5] K.Huarng, “Heuristic models of fuzzy time series for forecasting”, Fuzzy Sets and Systems, Vol. 123, pp 369-386, 2002.
[6] K. Huarng, “Effective lengths of intervals to improve forecasting in fuzzy time series”, Fuzzy Sets and Systems, Vol.12,
pp.387-394, 2001.
[7] C.C. Hsu and S.M. Chen, “ A new method for forecasting enrollments based on fuzzy time series”, Proceedings of the
Seventh Conference on Artificial Intelligence and Applications, Taichung, Taiwan, Republic of China, pp. 17-22.
[8] J.R. Hwang, S.M.Chen, C.H.Lee, “Handling forecasting problems using fuzzy time series”, Fuzzy Sets and Systems,
Vol. 100, pp,217-228, 1998..
[9] T.A.Jilani and S.M.A. Burney, “M-factor high order fuzzy time series forecasting for road accident data”, In IEEE-IFSA
2007, World Congress, Cancun, Mexico, June 18-21, Forthcoming in Book series Advances in Soft Computing,
Springer-Verlag, 2007.
[10] T.A. Jilani, S.M.A. Burney and C. Ardil, “Multivariate high order fuzzy time series forecasting for car road accidents:,
International Journal of Computational Intelligence, Vol. 4, No.1, pp.. 15-20., 2007.
[11] G..J. Klir and T.A.Folger, Fuzzy Sets, Uncertainty, and Information, Prentice Hall, New Jersey, U.S.A, 1988.
[12] G.J. Klir and B.Yuan, Fuzzy Sets and Fuzzy Logic: Theory and Applications, Prentice Hall, New Jersey, U.S.A., 2005..
[13] L.W. Lee, L.W.Wang and S.M.Chen, “Handling forecasting problems based on two-factors high-order time series”,
IEEE Transactions on Fuzzy Systems, Vol. 14, No.3, pp. 468-477, 2006.
[14] H.Li, R. Kozma, “A dynamic neural network method for time series prediction using the KIII model”, Proceedings of the
2003 International Joint Conference on Neural Networks, 1: 347-352, 2003.
[15] S. Melike and Y.D.Konstsntin, “Forecasting enrollment model based on first-order fuzzy time series”, in proc.
International Conference on computation al Intelligence, Istanbul, Turkey, 2004.
[16] Q.Song, “A note on fuzzy time series model selection with sample auto correlation functions”, Cybernetics and Systems:
An International Journal, Vol. 34, pp.93-107, 2003.
[17] Q.Song and B.S.Chissom, “Fuzzy time series and its models”, Fuzzy Sets and Systems, Vol. 54, pp, 269-277, 1993.
[18] Q.Song and B.S.Chissom, “Forecasting enrollments with fuzzy time series part I”, Fuzzy Sets and Systems, 54: 1-9.
[19] Q.Song and B.S.Chissom, “Forecasting enrollments with fuzzy time series: Part II”, Fuzzy Sets and Systems, Vol. 62:
pp. 1-8, 1994.
[20] Q.Song and R.P. Lcland, “Adaptive learning defuzzification techniques and applications:, Fuzzy Sets and Systems, Vol.
81, pp.321-329, 1996.
[21] S.F.S and, S.H. Li, “Neural network based fusion of global and local information in predicting time series”, Proceedings
of the 2003 IEEE International Joint Conference on Systems, Man and Cybernetics,
No. 5: pp. 4445-4450, 2003.
[22] J. Sullivan and W.H. Woodall, “ A comparison of fuzzy forecasting and Markov modeling”, Fuzzy Sets and Systems,
Vol.64, pp.279-293, 1994.
[23] L.X.Wang and J.M.Mendel, “Generating fuzzy rules by learning from example”, IEEE Transactions on Systems, Man,
and Cybernetics, Vol.22, pp.1414-1427, 1992.
61
AUTHORS PROFILE
Arutchelvan Govindarajan received B.Sc degree in Mathematics from University of Madras, Chennai, M.Sc., degree in
Mathematics from Bharathidasan University, Trichy, M.Ed., degree in Mathematics from Annamalai University,
Chidambaram, M.C.A., degree in Computer Applications from Madurai Kamaraj University, Madurai, M.Phil., degree in
Computer Science from Manonmaniam Sundaranar University, Tirunelveli, PGDCA in Computer Application & DSADP in
Computer Science from Annamalai University, Chidambaram. Presently He is doing Ph.D in Computer Science at SCSVMV
University, Kancheepuram. He is the Vice-Principal & Head of the Department of Computer Science and Applications at
Adhiparasakthi College of Arts & Science, G.B.Nagar, Kalavai, Vellore Dist., Tamil Nadu . He has 22 years teaching
experience in Computer Science and Applications. His research interest is in the area of Fuzzy Logic, Artificial Intelligence,
Artificial Neural Networks. He is also a member of Computer Society of India.
Dr. S.K.Srivatsa
Senior professor, St. Joseph’s College of Engg,
Jeppiaar Nagar, Chennai-600 064
Dr.S.K.Srivatsa was born at Bangalore on 21stJuly 1945. He received his Bachelor of Electronics and Communication
Engineering Degree (Honors) from Javadpur University (Securing First Rank and Two Medals). Master Degree in Electrical
Communication Engineering (With Distinction) from Indian Institute of Science and Ph.D also from Indian Institute of
Science, Bangalore. In July 2005, he retired as professor of Electronics Engineering from Anna University. He has taught
twenty-two different courses at the U.G. Level and 40 different courses at P.G. Level during the last 32 years. He has
functioned as a Member of the Board of Studies in some Educational Institutions. His name is included in the Computer
Society of India database of Resource Professionals. He has received about a dozen awards. He has produced 23 PhDs. He is
the author of well over 350 publications.
Dr.Jagannathan Ramaswamy received B.Sc, M.Sc and Ph.D degrees from the University of Madras, India He obtained his
Master of Philosophy degree in Space Physics from Anna University, Chennai. He was the Reader and the Head of the
Postgraduate Department of Physics at D.G.Vaishnav College, Chennai.
Dr.Jagannathan is currently the Chairman cum Secretary of India Society of Engineers, Madras Chapter, Managing Editor
(Publications), Asian Journal of Physics and Deputy Registrar (Education), Vinayaka Missions University, Chennai, India.
62
Recognition of Divergent Behaviour of a Plant Process Based on Direction of

Deviation of the Patterns Using Artificial Neural Networks
Syeda Asra
Department of Computer Science & Engineering
Appa Institute of Engineering & Technology, Gulbarga, India
asrasyedaali@yahoo.co.in
Abstract
The proposed system performs grouping of temporal patterns. Temporal patterns involve ordered sequences of data
appearing in time.The goal of this paper is to automate the recognition of the divergent behaviour of the plant process by
combining the patterns based on the characteristics of the patterns using neural network and to assign them to one of the
prespecified class also called category.Many times it is required to quickly check a direction of deviation of a pattern in order
to find out the divergent behaviour of a new plant process for quick action to be taken.The proposed system best suits for this
scenario since here high speed learning is achieved.Less emphasis is given on nature of deviation of the patterns. Patterns
were randomly combined and neural network learnt the behaviour of only those combination of patterns which shared some
common features.The network learnt and have given the outstanding result.This set up can be used to find the direction of
deviation of the new pattern to find out performance of the plant that is if it is performing above some threshold limit or
below some threshold limit or normal.In either case the action taken will be certainly different.Hence this system acts as a
decision support for the control unit to take appropriate action.In complex processes were there are large amount of patterns
training individual patterns takes a longer time,and finding the relationship amongst them is a critical task.In the proposed
system high speed learnig is achieved.This is an automated system which helps you do that.After a lot of research the
conclusion is that any pattern can have one of the three types of patterns.They are increasing,decreasing and lastly
normal.Control chart provides the information of flow about patterns.The proposed system works on core methodology of
neural network to find the deviation from normal and is not an application specific but rather a generalized model based on
simulating the data.
Keywords: Control Chart, Simulation, Deviation, Back Propagation, ANN.
63
1. Introduction
Artificial neural network are massively parallel-interconnected networks of simple adaptive elements and their
hierarchical organizations, which are intended to interact with the objects of the real world in the same way as the biological
counterparts. Neural networks find wide applications in parallel distributed processing and in real-time environments.
Neural Networks have considerable advantages over expert systems [1] in terms of knowledge acquisition, addition of new
knowledge, performance and speed. Recently, interest in the application of associative memories and neural networks to
problems encountered in diagnostic expert systems development has increased. Neural Networks appear to offer features,
which coincide well with the requirements of pattern-based diagnosis. An important feature of neural networks is that they
can interpolate the training data to give an appropriate response for cases described by neighboring or noisy input data [2].
This paper describes the design and simulation of a neural network for categorizing the patterns based on the characteristics.
1.1 Neuro Computing

Normally the goal of pattern classification is to assign a physical object, event or phenomenon to one of the pre
specified classes. However, in the proposed system instead of classifying [3] a single pattern the system group’s two patterns,
which inherently, share some common characteristics. This capability of neural network helps to find the divergent behavior
of plant process. These patterns combined have been given the same target. Out of all possible combinations only two sets of
combination was possible. The results are outstanding.
2. Methodology
Neural networks are computational models that share some of the properties of the brain. These networks consist of
many simple “units” working in parallel with no central control, and learning takes place by modifying the weights between
connections. The basic components of an ANN are “neurons”, weights, and learning rules. In general, neural networks are
utilized to establish a relationship between a set of inputs and a set of outputs. Figure 1 shows the general architecture of a
multilayer, feed forward neural network. ANN’s are made up of three different types of “neurons”. (1) input neurons (2)
hidden neurons and (3) output neurons. Inputs are provided to the input neurons, such as machine parameters, and outputs are
provided to the output neurons. These outputs may be a measurement of the performance of the process, such as part
measurements. The network is trained by establishing the weighted connections between the input neurons and output
neurons via the hidden neurons. Weights are continuously modified until the neural network is able to predict the outputs
from the given set of inputs within an acceptable user-defined error level.
Figure 1: Multilayer Feed Forward Artificial Neural Network
The activation function used during training is the sigmoid function defined in Equation 1.
(1)
Initially, weights connecting input and hidden units and weights connecting hidden and output units are assigned random
value:
(2)
(3)
64
Where w1ij = weights between unit i in input layer to unit j in hidden layer w2ij = weights between unit i in hidden layer to
unit j in output layer
(4)
(5)
Activation values for units in the input layer to units in the hidden layer are calculated by
(6)
where xi = value of unit i in the input layer and xo = 1.0 hi activation value of unit i in the hidden layer
Activation from units in the hidden layer to units in the output layer are propagated using:
((( (7)
where oj = activation value of unit j in the output layer Errors, δj, in the units in the output layer, are computed using
(8)
where yj = target value of unit j in the output layer Errors in the units in the hidden layer, represented as δ1j, are determined
by
(9)
Weights are adjusted between the hidden layer and output layer (Equation 10) and between the input and
hidden layer (Equation 11) where the learning rate is denoted h;
(10)
(11)
Once the network is sufficiently “trained”, a general model is created for the relationship between inputs and outputs. The
user can then determine the impact that a specific set of process inputs has on a set of process outputs. ANN’s need to know
little about the process itself; therefore, a network can be constructed and trained with a set of representative data from the
process. A network has the capability of learning and adapting to situations, which it has not yet seen due to its generalization
capabilities. After sufficient training, the ANN model can be used to monitor new patterns to determine, with the current
patterns, if the process is moving toward an “out of control” situation in the sense deviating from the normal than it is
increasing or decreasing. The advantages of using an ANN for process control are overwhelming and in many cases can
greatly help the control unit to take appropriate decision.
3. Simulation of Data
Control chart gives the information of flow about the patterns.The proposed system works on simulated data.After a lot
of research, the outcome is that there can only be maximum six types of variation in the temporal patterns as shown in Fig
65
2.Starting from top left to right they are 1)Normal 2)Cyclic 3)Incremental 4)Decremental 5)Upward Shift 6) Downward
Shift.Normal pattern indicates the plant is working in a pefect condition.Cyclic indicates almost normal but with few
harmonics.Incremental indicates the patterns behaviour increasing,decremental indicates patterns bevaiour
decreasing.Upward shift indicates shift in the pattern in the upward direction after behaving normally for some
time.Downward shift indicates shift in the pattern in the downward direction after behaving normally for some time.Hence at
sixty different points the pattern has been sampled.We have input layer consisting of 60 so that maximum features can be
extracted.These patterns cannot be directly trained, since the network is trained by establishing the weighted connections
between the input neurons and output neurons via the hidden neurons. These weights are some random numbers.So there is a
need to simulate numeric data for these patterns.500 sets of data is generated using the formulae.
Figure 2: Control Chart of Six Patterns
3.1 Formulae for Data Simulation

1. Normal pattern
P (t) =   r (t)  
2. Cyclic pattern
P (t) =   r (t)   + a  sin  2  t / T)
3. Increasing trend
P (t) =   r (t)   + g  t
4. Decreasing trend
P (t) =   r (t)    g  t
5. Upward shift
P (t) =   r (t)   + b  s
6. Downward shift
P (t) =   r (t)    b  s
where
 Nominal value if process variable
 Standard deviation of process variable
a amplitude of cyclic variation in cyclic
pattern
g gradient of increasing /decreasing trend
pattern
b shift position in up/downward shift pattern
smagnitude of shift
r ( ) random number
T  period of cycle in cyclic pattern
P (t)  value of sample data point at time t
66
3.2 Normalization
Once the data is generated, the data may vary from each other by a large amount. This would result in the machine to
take a long time to learn. Thus, normalization is done in order to allow the data to vary between 0 &1. There are different
methods in which the data can be normalized. One among them is the following.
1) Finding the maximum value from each input data
2) Dividing each of the value of the input data with the maximum value
4. An overview of the proposed system

The framework developed in this research paper is a closed-loop artificial neural network system for automated
recognition of deviation of the patterns. The system consists of back propagation neural networks, with the functional block
diagram depicted in Figure 3. The input nodes are 60, hidden are 10 the output nodes are 3.
The network is presented with every different pair of patterns .Each pair is given a target. The network is trained with all
possible combinations of the patterns. The network learns the behavior of only the combination of 1) Normal-Cyclic 2)
Incremental-Upward Shift 3) Decremental-Downward shift. Here grouping the normal with cyclic is of no importance since
it does not give the direction of deviation. The normal pattern can be considered as an indication of the plant working
satisfactorily as expected. Data is generated for 500 blocks of data. Training took place using 10 data sets and testing with
490 data sets. During testing the patterns which the network has not been trained for is given. This proves that the ANN is
having a capability of generalization. Thus, during testing when an input is given as cyclic or normal it gives the output as
one. Similarly, when the input is given as incremental or upward shift the output is a one, and same follows for upward shift
and incremental pattern.
Thus, my research work about artificial neural network infers that the ANN groups the patterns having similar
characteristic. This property of ANN could be utilized in situation where the deviation from normal is concerned rather than
precisely detecting the signals. Fig 4 shows the error plot.
5. Neural Network Testing & Training

The training is give offline but can be used online due to its high-speed learning capability. The feedback neural
network is a back propagation, supervised-learning network. The network selected had 10 hidden layers, in addition to 60
input layers and 3-output layer. The targets for normal-cyclic, incremental upward shift & decremental downward shift
was [1 0 0], [0 1 0] & [0 0 1] respectively. Network training was performed on 10 data sets and testing with 490 data sets.
The learning rate was 0.9. Experimentation with bias node was done. The RMS error [3] of the test data was observed to be
 0.00197 from figure 4.
Figure 3: Block Diagram of Proposed System
67
Figure 4: Error plot for Normal-Cyclic

6. Results
After the ANN have been trained with 10 blocks of data, by combining all possibilities. Finally it is concluded that the
network learns only the behavior of the combination of normal-cyclic, incremental-upward shift, decremental-downward shift
.A test is performed.ANN is presented with either normal or cyclic pattern as input, it recognizes an accuracy as high as 99%
is achieved. The same follows for other combinations. From figure 5, it is obvious that the network learnt the behavior of
normal cyclic but it is of not much use for us since we are interested in finding the deviation of the patterns. This is just to
show normal and cyclic can be trained.
Figure 5: Results of Normal-Cyclic
Figure 6: Results of Incremental-Upward shift
Figure 7: Results of Decremental-Downward shift
68
Table 1: Performance Evaluation
Efficiency in % Execution time in

mins
With Without With bias Without
bias bias bias
98.75% 99.3333% 1.2333 1.0120
Table 2: Performance Analysis for Different combination of Input Patterns
Name of the Combination Eff in Exec

% Time
in
mins
Incremental- Upward shift
Decremental – Downward
99.91 1.01
shift
Normal-Incremental
Upward – Decremental 35.91 1.24
Downward shift- Cyclic
Upward-Cyclic
Normal-Decremental
Downward shift 39.08 0.56
Incremental
Normal-Downward shift
Cyclic-Upward shift 44.23 0.73
Incremental –Decremental
8.Conclusion and Future Research

After carrying out this project, conclusion is that network never learns the behaviors of dissimilar patterns and it only
learns and later recognizes the patterns which have inherent some common features. Just like humans, we like to be in
company of someone, which matches to our wavelength. So does ANN. Observing the execution time taken, it is obvious
that Neuro system designed is well suited for the real time applications where it can be used online. This proposed system can
be used as partial neuro monitoring system, where only direction of deviation is important not the nature of deviation. High-
speed learning is achieved. This project is implemented in Gradient descent rule. It can be extended by implementing in
Genetic algorithms. John Holland invented genetic algorithms in 1960. Holland’s original goal was not to design algorithms
to solve specific problems, but rather formulate the phenomenon of adaptation as it occurs in nature & develop ways in which
the mechanism of natural adaptation might be imitated into computer systems. Holland’s GA is a method of moving from one
population of chromosomes to a new population for using a kind of ‘natural selection’ together with genetics inspired
operators of crossover, mutations & inversions. GA‘s work by discovering, emphasizing and recombining good building
blocks of solutions in a highly parallel fashion. The idea here is that good solutions tend to be made of good building blocks –
combination of bit values that confer higher fitness on the strings in which they are present. Holland introduced the notion of
Schema’s to formalize the informal notion of building blocks.
69
ACKNOWLEDGEMENT
The author would like to acknowledge supervisor Dr. M.K Singh from IISc Bangalore for his unlimited support, without
his notes and suggestion this research would not appear.
References
[1] Regina Barzilayy, Daryl McCullough(1999) “A New Approach to Expert System Explanation”
[2] Lambert Spaanenburg, “Early Detection of Abnormal Emergent Behaviour”
[3] M Khoshnevisan, “A Proposed Artificial Neural Network Classifier to Identify Tumor Metastases”
Mr. Timothy James Stich, Dr. Julie K. Spoerre & Dr. Tomas Velasco. (1999)”A Journal of industrial technology vol
16,pp1.
[4] Jacek M.Zurada, “An Introduction to Artificial Neural System”
70
Application of Fuzzy Priority Scheduler for supporting QoS in MANET for DSR
Protocol
Prof.S.P. Setti Vijay Kumar D V and Nagedra Prasad G S M
CS&SE Dept., AU College of Engineering, Andhra University, India
drspsetty@yahoo.com vijaykdasari@yahoo.com chakri.naresh@gmal.com
Abstract
A wireless Ad-hoc network consists of wireless nodes communicating without the need for centralized administration, in
which all nodes potentially contribute to the routing process. In this paper, we analyze packet scheduling algorithm to find
those that most improve performance in congested network. Hence, a scheduling algorithm to schedule the packet based on
their priorities will improve the performance of the network. Packet schedulers in wireless ad hoc networks serve data
packets in FIFO order. Here, we present a fuzzy based priority scheduler for mobile ad-hoc networks, to determine the
priority of the packets using Dynamic Source Routing (DSR) as the routing protocol. The performance of this has been
studied using QUALNET5.0 simulator and measured such as packet delivery ratio, average end-to-end delay and throughput.
It is found that the scheduler provides overall improvement in the performance of the system.
Keywords: MATLAB fuzzy tool, Fuzzy Priority Scheduler, QUALNET5.0, DSR, MANETs.
71
1. Introduction
A collection of nodes that communicate with each other by forming a multihop radio network and maintaining connectivity
in a decentralized manner is called an ad hoc network. In these networks, the mobility of nodes and the error prone nature of
the wireless medium poses many challenges like frequent route changes and packet losses. Such problem increases packet
delay and decreases throughput. To improve the performance and maintain the QoS in MANET the scheduler can be used. C.
Gomathy and S. Shanmugavel [1] and [2] has been design a fuzzy based priority scheduler to determine the priority of the
packets. Based on this paper, we are trying to explore how to achieve the better performance of end-to-end QoS target in
MANETs.
In this paper, to provide a fuzzy based priority scheduler, this improves the QoS parameters in wireless ad hoc
network. Hence, introducing a scheduling algorithm to determine which queued packet to process next will improve the
overall end-to-end performance. Without scheduling, the packets will be processed in FIFO manner and hence there is more
chance that either more packets may be dropped or may not meet the QoS target. A scheduler should schedule the packets to
reach the destination quickly, which are at the verge of expiry. To incorporate the scheduler for the existing routing protocols,
various routing protocols are studied [3]. It is found that the proposed fuzzy scheduler provides improved packet delivery
ratio, reduced end-to-end delay and increased throughput.
2. SCHEDULING SCHEMES FOR PROTOCOLS
2.1 DSR PROTOCOL
DSR is an on-demand, source routing protocol. Transmitting nodes discover the route to their destination nodes on
demand. This route is included in the data packets as the route header.
The DSR protocol consists of two phases – Route Discovery and Route Maintenance. When a source node A wants to send
a packet to a destination node B, it first checks if it already has a route to B stored in its cache. If the route is not stored, a
Route Request (RREQ) packet is broadcast with the address of node A in the route record. An intermediate receiving node
checks if its route cache has a route to the destination node. In that case, it appends the route in the route record and sends
back a Route Reply (RREP) packet by using the reverse route (assuming symmetrical links). If the intermediate receiving
node does not know the route, it appends its own address to the route record and broadcasts another RREQ packet. Using the
route cache helps conserve network resources by limiting the number of route discovery packets. It is possible that a node
receives route request packets for the same source-destination pair but with different route headers that represent different
routes. In this case, the node chooses the route with the lowest cost, usually the shortest path.
When a destination node receives the RREQ packet, it then appends its address to the route record and sends back an
RREP packet. If the links are symmetric, it constructs the route to the source by reversing the route in the received route
record. If asymmetric, the destination initiates a route discovery to the source and receives the RREP packet upon its
successful completion. When the source begins transmitting data packets that are addressed to the destination node through
the discovered route, any intermediate receiving node caches the route to the destination node, if the node has not cached
already. The details of other optimizations can be found[4].
In the Route Maintenance phase, if a transmitting node encounters a fatal error or does not receive acknowledgments of
the transmitted packets from the downstream node, it generates a Route Error (RERR) packet. It also removes the route(s)
that use this failed link from its cache. Furthermore, nodes between this node and the source remove the route(s) with the
reported failed link from their caches upon receipt of the RERR packet. When the RERR packet makes its way to the source,
a new route discovery process may be initiated.
2.2 SCHEDULING SCHEMES

For improving the performance of the mobile ad-hoc networks, a scheduler can be used. There are several scheduling
policies for different network scenarios. In DSR, the remaining hops to a destination are stored in own routing table, but the
entries in this list may change dynamically over time, so the advertisement must be made often enough to ensure that every
mobile computer can almost locate every other mobile computer. The drop tail policy is used as queue management
algorithm in all scheduling algorithm. It drops packets from the tail of the queue when channel capacity is full. Except for the
no priority scheduling algorithm, all the other scheduling algorithms give higher priority to control packets than to data
packets. In priority scheduling, control and data packets are maintained in separate queues in FIFO order and it gives high
priority to control packtes. Currently, only this scheme is used in mobile ad-hoc networks [5]. Considering the suitability of
72
the different types of scheduling methods for MANET, several scheduling schemes have been proposed[6]. Since the three
variables viz., expiry time of packet, queue length and data rate are considered as input variables, using the fuzzy logic to find
the priority index of the packet is found to be suitable in improving the overall performance of MANETs.
3. THE FUZZY SCHEDULER

Fuzzy logic implements human experiences and preferences via membership functions and fuzzy rules. The fuzzy
scheduler proposed here, calculates the priority index of each packet. The fuzzy scheduler uses three input variables and one
output variable. The three input variables to be fuzzified are the expiry time and data rate of the packet and Queue length of
the nodes to which the packet is associated with. The inputs are fuzzified, implicated, aggregated and defuzzified to get the
crisp value of the output.
The linguistic variables associated with the input variables are low (L), medium (M), and high (H). For the output
variable, priority index, five linguistic variables are used. They are very low (VL), low (L), medium (M), high (H) and very
high (VH).
The table 1 shows the fuzzy conditional rules for the fuzzy scheduler. The three input variables have 27 combinations
(3*3*3) and the corresponding output is shown in the tabulation. The rule base is split into three tables and the first table
gives out the rule base for Expiry time low and nine combinations of the other two input variables. The Table gives out the
rule base for Expiry time medium and the third for expiry time high. To illustrate one rule in the first table, the first rule can
be interpreted as, if (Expiry time is low) and (Date rate is low) and (Queue length is low), then priority index is low. Since in
this rule, Data rate and Queue length are low and packets are associated with low delay, the priority index is set to be low. In
the Table, for medium expiry time when data rate and queue length both are high, the priority index is set to be medium as
seen from the last column of the table. Similarly the other rules are framed.
The output priority index, if very low, indicates that packets are attached with a very high priority and should be
immediately scheduled. Similarly, if the priority index is very high, it indicates that packets are attached with least priority
and will be scheduled only after all high priority packets are scheduled [9] and [10].
Table 1: Fuzzy Rule Base

D/Q L M H
Expiry time (low)

L L L VL
M VL VL VL
H L VL VL
Expiry time (medium)
L M M L
M M M L
H M M M
Expiry time (high)
L VH VH H
M H M M
H H H M
4. PERFORMANCE EVALUATION
The simulation for evaluating the fuzzy scheduler was implemented within the QualNet library. The simulation package
QualNet version 5.0 is used to analyze and evaluate the performance of the proposed fuzzy scheduler. The QualNet version
5.0 [7] and [8], a software that provides scalable simulations of Wireless Networks. In this simulation, we consider a network
of 20, 40, 60, 80, 120 and 150 nodes (one source and one destination) placed in a dimension of 1000m x 1000m area. Each
simulation is run for 600 seconds of simulation time. Multiple simulations run with different seed values were conducted for
each scenario and collected data was averaged over those runs. Table 2 lists the simulation parameters.
73
4.1 Performance Metrics

The following metrics are used to evaluate the effect of fuzzy scheduler. The metrics were derived from one suggested by the
MANET working group for routing protocol evaluation.
Table 2: Simulation Parameters
Simulation Environment
Area 1000m x 1000m
Simulation Time 600 Sec
Nodes 20, 40, 60, 80, 120, 150
Nodes Placement Random
Path loss Model Two Ray
Mobility Model Random Way Point
Pause Time 30 Sec
Maximum Speed 10mps
Traffic CBR
Packet Size 512 bytes
MAC layer 802.11
Average end-to-end delay: End-to-end delay indicates how long it took for a packet to travel from the source to the
application layer of the destination.
Packet Delivery Ratio: Packet delivery ratio is the ratio of the number of data packets actually delivered to the destination to
the number of data packets supposed to be received. This number presents the effectiveness of the protocol.
Throughput: This is measured in bytes per sec, which also serve as the performance measure for the scheduler.
Figure 1: Average End-to-End with No. of Nodes
74
Figure 2: Packet Delivery Ratio with No. of Nodes
Figure 3: Throughput with No. of Nodes
5. Conclusion and Future Scope
In this paper, we have analyzed the performance of the fuzzy-based priority scheduler for data traffic and evaluated the effect
of inclusion of this scheduler with underlying multicast routing protocol DSR. We have addressed a fuzzy-based priority
scheduler for data packets, which improves the quality-of-service parameters in mobile ad hoc networks. The fuzzy scheduler
attaches a priority index to each packet in the queue of the node. Unlike the normal sorting procedure for scheduling packet, a
crisp priority index is calculated based on the inputs such as queue length, data rate, and expiry time of packets, which are
derived from the network. The membership functions and rule bases of the fuzzy scheduler are carefully designed and the
output is verified using Matlab7.0 fuzzy logic toolbox with FIS editor. Then the inputs are identified in the library of
Qualnet5.0 and the fuzzy scheduler is attached. This work can be also extended with inclusion of this fuzzy scheduler with
different underlying multicast routing protocols with different mobility models.
References
[1] C. Gomathy and S. Shanmugavel, “Supporting Qos in manet by a fuzzy priority scheduler and performance analysis with
multicast routing protocols” in EURASIP journal on wireless communications and networking, 2005:3, 426-436.
[2] C. Gomathy and S. Shanmugavel, “Performance evaluation of a novel Fuzzy based Priority Scheduler formobile AdHoc
networks and its effect onMAC protocols,” in Proc. 12th International Conference on Advanced Computing &
Communication (ADCOM ’04), Ahmedabad, Gujarat, India, December 2004, at IEEE Gujarat section.
[3] Royer, E.M., 1999. “A review of current routing protocols for ad hoc mobile wireless networks”.IEEE Personal
Communications, pp: 46-55.
[4] David A. Maltz, Josh Broch, Jorjeta Jetcheva, and David B. Johnson. “The Effects of On-Demand Behavior in Routing
Protocols forMulti-Hop Wireless Ad Hoc Networks”. IEEEJournal on Selected Areas in Communicationsspecial issue on
mobile and wireless networks, August 1999.
75
[5] D.John.son and D.Mdtz. “Mobile Computing”. Chapter on Dynamic source routing in Ad-hoc wireless networks, Kltiwcr
Academic, 1996.
[6] V. Kanodia, C. Li, A. Sabharwal, B. Sadeghi, and E. Knightly, “Distributed priority scheduling and medium access in ad
hoc networks,”Wireless Networks, vol. 8, no. 5, pp. 455–466, 2002.
[7] QualNet Network Simulator; Available: http://www.scalable-networks.com.
[8] QualNet documentation, “QualNet 5.0 Model Library: Advanced Wireless”; Available: http://www.scalable-
networks.com/products/qualnet/download.php#docs.
[9] Matlab 7.0; Available: http://www.mathworks.com.
[10] Matlab documentation, Available: MATLAB701/help/techdoc.
76
The Roll of Knowledge Shearing and Web-Based Applications on Learning

Language
Hamsa H. Aljudi 1, B.B.Zaidan 2, A.A.Zaidan 2, Zahraa M.Hamad 1, Rami Alnaqeib 3,

Hayan.T.Madhloom3
1
Department of English Language Study / Faculty of Social Science and Humanities School of
Language Studies and Linguistics / University Kebangsaan Malaysia / Malaysia
2
Department of Electrical and Computer Engineering, Faculty of Engineering, Multimedia University,
63100 Cyberjaya, Selangor Darul Ehsan, Malaysia
3
Faculty of Computer Science and Information Technology / University of Malaya / Kuala
Lumpur/Malaysia
Email Corresponding Auther: aws.alaa@gmail.com
Abstract
Social networking website has booked the people by hurricane the way society connects and interacts with each other.
As the world is starting to catch up with the existing paid or free social networking sites, however, nothing so far is catered
for the purpose of shearing knowledge to be side by side with exist social network. Knowledge shearing can play an
important role on enhancing the skill of writing and learning language, however, there is negative knowledge sheared in
social network, this negative knowledge may affect badly on the main idea of the knowledge shearing. As result knowledge
there is not much article talk about the negative learning for undirected learning for using social sites
Keyword: - Knowledge Shearing, Social Network, Computer Mediated Communication, learning Language
77
1. Introduction
Knowledge management is the explicit and systematic management of vital knowledge and its associated processes of
creating, gathering, organizing, diffusion, use and exploitation. Knowledge sharing is an activity through which knowledge
such as information, skills, or expertise sheared among people from different level of education and expertise for the purpose
of improving the skills of less knowledge in a Particular topic [1, 2].
For many reasons some teachers would want to integrate computers into their classrooms. First, it Stimulate students to
interact more in their target language (TL) [3-5]. In other word, CMC can facilitate learners to interact with other speakers of
the TL without being in close geographical closeness. Furthermore, students feel less anxious about using the TL by CMC [6,
7] and so they can reflects themselves more freely, contentedly, and creatively during CMC and they added that CMC helped
increase students’ thinking ability, and students felt less stressed while communicating via CMC. The concept of knowledge
sharing is used frequently in many studies [8, 9] the goal of knowledge sharing is either explore new knowledge, skill or
exploit an existing knowledge [10, 11]. Knowledge shearing consists of Implicit and Explicit knowledge; implicit knowledge
is the knowledge that learners acquire and use unconsciously, and explicit knowledge is knowledge that requires conscious
teaching and learning.
2. Type of Knowledge
Many type of knowledge has been reported in literature, which is as fallow:
A. Knowledge transfer
Knowledge transfer: is the sensible problem of transferring knowledge from one part of the organization to another part
of the organization. This kind of knowledge is widely found in the fields of organizational development and organizational
learning.[12, 13]. In other word, Knowledge transfer is the process through which one network member is affected by the
experience of another [12].

Figure 1: depicted knowledge transfer
78
B. Knowledge Market
A knowledge market is a mechanism for distributing knowledge resources, for example, buyers and sellers of knowledge
negotiate a mutually satisfactory price for the knowledge exchanged [14]
Figure 2: depicted knowledge market

3. Computer Mediated Communication
Computer mediated communication is a field of computer applications to education and training that is rapidly
expanding both in terms of the sheer number of practical applications [15]. While the term has traditionally referred to those
communications that occur via computer mediated formats, for instant messages, e-mails, and chat rooms; it has also been
applied to other forms of text-based interaction such as text messaging [16]. By the turn of the century the Web had entered
most aspects of our lives from communications to E-Government, E-Commerce and E-Learning, making it much more than
just an information repository. Nowadays, computers are being used for research on language use not only in classroom
settings, but also in the context of e-mail and internet chat as it has become more extensive. Interaction between two or more
speakers using online chat (i.e., synchronous CMC) has established a substantial amount of attention as well. With the
beginning of CMC, it has become a simple and cheap matter to create language learning groups all over a country and around
the world[17]. Yet if CMC cannot be used in a classroom setting (due to local shortages of computers or networking
technology, for example), L2 learners can still be exposed to and use CMC via the Internet.
4. The Affect of Social Network on the Writing Apprehension
Computers have become an accepted tool in writing classes, and research on computers and various aspects of the
composing process has mushroomed in the last decade. [18]. In the past few decades, writing and technology have focused on
a variety of computer applications as well as a variety of tools, from word processors, e-mail, list serves, to online chats,
bulletin board discussions, and Web page projects [19]. Many educators have used blogs in their classes, they have also
written about those experiences and how blogging is grounded in current learning theory (e.g., process, post-process, and
genre). Additionally, blogs have been used to demonstrate other theoretical and pedagogical frameworks in both academic
and professional settings[19].
79
Figure 3 shows a caption from a website talking about the role of blogs on enhance the writing
5. Negatives Knowledge Shearing
In our case, we are looking forward to enhance the writing skills using the knowledge shearing in the social science
sites, blogs and chatting. In fact, most of the young people texting each other using “text messaging and chat abbreviations”,
according to webopedia.com, they collect hundreds or maybe thousand of short form messages, the example below showing
the development of the short “texting using any blogs or chatting room”. It is noticeable that some of thirteen’s or fourteens
would not understand these messages form. Furthermore, there will be a bad impact on the spelling and academic writing;
however, we can not neglect the role of knowledge shearing in the part of learning .
Example
Two friends, X and Y use to chat, the dialog become as fallow:
X: hi Y: Hi
X: What's up? Y: Nothing too much
Y: Got to go
Y: I am sorry
X: Never Mind
X: Take your time
X: Have a nice day Y: U2
The above conversation is the old type of chatting, while the figure below showing the modern chatting language
80
The above conversation is the old type of chatting, while the figure below showing the modern chatting language
X: HI X: HI
Y: HI Y: HI
X: ZUP!! X: ZUP!!
Y: N2M Y: N2M
Y: G2G Y: G2G
Y: IMS Y: IMS
X: NVM X: NVM
X: TYT X: TYT
X: HAND X: HAND
Y: U2 Y: U2
Figure 4: showing the modern chatting words form.
6. Conclusion
Knowledge shearing has played a significant role to improve the skill of writing and learning language for L2 writers,
the process of writing in the academic environment represent a challenge. CMC can facilitate learners to interact with other
speakers of the TL without being in close geographical closeness. The literature reported many research suggest and they
positively recommend using blogs and social sites to improve the skill of academic writing, unlike the literature, the author
has present a daily example of texting chat; in this example the author has shown the modern chatting texting “chatting
language”. This language is almost new writing language, after four or five months, the users will lost many skill of writing.
7. Acknowledgement
This project has been partially funded from University Kebangsaan Malaysia, Department of English Language Studies.
8. References
1. Hansen, M.T., The search-transfer problem: The role of weak ties in sharing knowledge across organization subunits.
Administrative science quarterly, 1999. 44(1): p. 82-85.
2. Wang, S. and R.A. Noe, Knowledge sharing: A review and directions for future research. Human Resource Management
Review, 2010. 20(2): p. 115-131.
3. Kern, R.G., Restructuring classroom interaction with networked computers: Effects on quantity and characteristics of
language production. Modern language journal, 1995. 79(4): p. 457-476.
4. Warschauer, M., Comparing face to face and electronic discussion in the second language classroom. Calico Journal,
1996. 13: p. 7-26.
5. Schwienhorst, K., Native-speaker/non-native-speaker discourse in the MOO: Topic negotiation and initiation in a
synchronous text-based environment. Computer Assisted Language Learning, 2004. 17(1): p. 35-50.
6. Hiltz, S.R., et al., The network nation. 1978: Addison-Wesley Reading, Mass.
7. Hiltz, S.R. and M. Turoff, The network nation: Human communication via computer. 1993: The MIT Press.
81
8. Wei, K. How Do Chinese Team Members Perceive the Impact of National Cultural Differences on Knowledge Sharing?
2007: Springer-Verlag New York Inc.
9. Wei, K., Sharing Knowledge in Global Virtual Teams. Virtuality and Virtualization, ed. I.I.F.f.I. Processing. Vol. 236.
2007: Springer Boston. 251-265.
10. Uzzi, B. and R. Lancaster, Relational embeddedness and learning: The case of bank loan managers and their clients.
Management Science, 2003. 49(4): p. 383-399.
11. Hargadon, A., How breakthroughs happen: The surprising truth about how companies innovate: Harvard Business
School Press. Cambridge, MA, 2003.
12. Inkpen, A.C. and E.W.K. Tsang, Social capital, networks, and knowledge transfer. Academy of Management Review,
2005. 30(1): p. 146-165.
13. Ghimisi, S. and G. Popescu, Centers tehnology transfer-active factor an the regional development. 2009, Munich
Personal RePEc Archive.
14. Ba, S., J. Stallaert, and A.B. Whinston, Research commentary: Introducing a third dimension in information systems
design--The case for incentive alignment. Information Systems Research, 2001. 12(3): p. 225-239.
15. Romiszowski, A.J., Computer mediated communication: a selected bibliography. 1992: Educational Technology Pubns.
16. Thurlow, C., L.B. Lengel, and A. Tomic, Computer mediated communication: Social interaction and the Internet. 2004:
Sage Publications Ltd.
17. Ingram, A.L., L.G. Hathorn, and A. Evans, Beyond chat on the internet. Computers & Education, 2000. 35(1): p. 21-35.
18. AYDIN, S., The Effect of Computers on the Test and Inter-Rater Reliability of Writing Tests of ESL Learners. The
Turkish Online Journal of Educational Technology TOJET 2006. 5(1): p. 7.
19. Jones, S.J., Blogging And ESL Writing: A Case Study Of How Students Responded To The Use Of Weblogs As A
Pedagogical Tool For The Writing Process Approach In A Community College ESL Writing Class, in Faculty of the
Graduate School. 2006, University of Texas at Austin: Austin. p. 322.
82
Behavioral model of 16 bit RISC Microcontroller in VHDL

Prof. Sanjay S. Chopade*1 Dr. Vijay Wadhai*2
*1
Asst. Prof. & HOD, SITRC, NASHIK
Dept. of Electronics & Tele.
sanjayschopade@rediffmail.com
*2
Professor and Dean, MAE Alandi
Dept. of Electronics
wadhai.vijay@gmail.com
Abstract
In this paper we have implemented 16 bit RISC machine. The RISC design gain its power by pushing some of less frequently
operations into software. Pipelining allows for simultaneous execution of parts or stages or instruction to process instruction
more efficiently. In this paper we have presented a simple method of behavioral of a RISC Microcontroller using VHDL. The
paper discusses the features of Microcontroller implemented using VHDL. The Experimental result of the instructions
designed in VHDL
Keywords: RISC, PIC16f84, VHDL
83
1. Introduction
In this paper we have implemented behavioral model of a 4 stage pipelined RISC based 16 Bit architecture of
microcontroller. The following equation is commonly used for expressing a computer's performance ability: [9]
The CISC approach attempts to minimize the number of instructions per program, sacrificing the number of cycles per
instruction. RISC does the opposite, reducing the cycles per instruction at the cost of the number of instructions per
program.[2].
Here we have designed 16 bit ALU, internal Memory, EPROM and all ports. All the features of PIC16F84-[3] has been
converted into 16 Bit and few additional instructions are design to enrich the instruction set.
2. Features of RISC Microcontroller
 On Chip ROM 8k x 16 bit

 On Chip RAM 256 x16 bit
 RISC BASED PROCESSOR
 On Chip File Registers 240 x 16 bit
 Special Function Hardware Register
 Bi-Directional 16 bit Port
 Logical, Boolean, Arithmetic Operations
 64K PHYSICAL SPACE for Memory.
 High performance RISC like CPU
 Only 37 instruction to learn
 16 – bit wide instruction.
 16 – bit address line
 16 -- bit wide data path
 Eight level deep hardware stack
 Direct, Indirect addressing modes for data and instructions.
 All instruction are single cycle except branch and literal instruction.
3. Architectural Design and Implementation

This Microcontroller is contains several major blocks as shown in the figure 3.1 Those are
RAM
ROM
CONTROL UNIT
PORTS
To begin with this controller, it uses a Harvard architecture in which program and data are accessed from separate memories
using separate buses. This improves bandwidth over traditional Von Neumann architecture where program and data are
fetched from the same memory using the same bus. Separating program and data busses further allows instructions to be
sized differently than 16.bit wide data word. Instruction op-codes are 16 bit wide making it possible to have all single word
instructions. A 16-bit wide program memory access bus fetches a 16 bit instruction in a single cycle.
A stage pipeline overlaps fetch and execution of instructions. Consequently, all instructions (37) execute in a single cycle
except for program branches and literal.
The Microcontroller address 8K x 16 program memory. All program memory is internal. The Microcontroller can directly or
indirectly address its registers files or data memory. All special function registers including the program counter are mapped
in the data memory. The MICROCONTROLLER has an orthogonal (symmetrical) instruction set that makes it possible to
carry out any operation on any register using any addressing mode. This symmetrical nature and lack of special optimum
‘situations' make programming with the MICROCONTROLLER simple yet efficient. In addition, the learning curves
reduced significantly.
84
The MICROCONTROLLER device contains an 16-bit ALU and working register. The ALU is a general purpose arithmetic
unit. It performs arithmetic and Boolean functions between data in the working register and any register file. The ALU is 16-
bit wide and capable of addition, subtraction, shift and logical operations. Unless otherwise mentioned, arithmetic operations
are 2’s complement in nature, two-operand instructions, typically one operand is the working register (W register). The other
operand is a file register or an immediate constant. In single operand instructions, the operand is either the W register or a file
register: The W register is an 16-bit working register used for ALU operations. It is not an addressable register, Depending on
the instruction executed, the ALU may affect the values of the Carry (C), and Zero (Z) bits in the STATUS register. The C
bits operate as a borrow out bit in subtraction. See the SUBLW and SUBWF instructions for examples. A simplified
Architecture of RISC Microcontroller block diagram for the MICROCONTROLLER is shown in Fig. 3.1
Figure 3.1: Architecture of RISC Microcontroller

3.1 Oscillator Block
The clock input (OSC1/CLKIN pin ) is internally divided by four to generate four non-overlapping quadrature clock namely
Q1, Q2, Q3 and Q4. internally, the program counter (PC) is incremented every Q1, and the instruction is fetched from
program memory and latched into the instruction register in Q4. It is decoded and executed during the following Q1 through
Q4. The clocks and instruction execution flow[3] is shown in Figure 3.2.
85
3.2 INSTRUCTION FLOW / PIPELINING

An Instruction Cycle consists of four Q cycles (Q1, Q2, Q3 and Q4). The instruction fetch and execute are pipelined such that
fetch take one instruction cycle while decode and executes will take another instruction cycle. However, due to the
pipelining, each instruction effectively executive in one cycle. If an instruction causes the program counter to change (e.g.
GOTO ) then two cycles are required to executive the instruction.
A fetch cycles begins with the program counter (PC) incremented in Q1. In the execution cycle, the fetched instruction is
latched into the instruction Register (IR) in cycle Q1. This instruction is then decoded and execute during the Q2, Q3, and Q4
cycles. Data memory is read during Q2 (operand read ) and written during Q4 (destination write).
Figure 3.2: Instruction Execution flow

3.3 Memory Organisation
Program memory Organization
This controller has a 16-bit program counter capable of addressing an 8K x 16 program memory space is physically
implemented. Accessing a location above the physically implemented address will cause a wraparound. The reset vector is at
0000H. Program memory organization shown in Figure 3.3
Figure 3.3: Memory Organization
Data Memory Organization
The data memory has only one bank which contain the general purpose registers and the special function registers. Bank
extends up to FFh (256 bytes). The lower locations of Bank are reserved for the Special Function Registers. Above the
86
Special Function Registers are General Purpose Registers implemented as static RAM. Data memory organization shown in
Figure 3.4
Figure 3.4: Data memory organization
3.4 GENERAL PURPOSE REGISTER FILE

The register file is accessed either directly, or indirectly through the file select register FSR. The general purpose register
locations are 0Fh-FFh.
3.5 STACK
This control has an 8 deep x 16-bit wide hardware stack .(see figure 3.5) The stack space is not part of either program or data
space, and the stack is not readable or writable. The PC is PUSHed in the stack when CALL instruction is executed. The
stack is POPed in the event of a RETLW instruction executed.
The stack operates as a circular buffer this mean that after the stack has been PUSHed eight times, the nineth push overwrite
the value that was stoered from the first push. The tenth push overwrite the second push (and so on).
Figure 3.5 8 deep X16-bit wide hardware stack
3.6 PROGRAM COUNTER

The Program Counter (PC) is a 16-bit register. PC is mapped in the data memory. PC is readable and writable just as is any
other register.
The 16-bit wide PC is incremented after each instruction fetch during Q1 unless:
Modified by GOTO, CALL, RETLW instruction Due to destination write to PC by an instruction
The operations of the PC for different instructions are as follows:
a) CALL instruction
b) GOTO instruction
c) Read instruction on PC
d) write instruction on PC
e) Read-modify-write instructions on PC
f) RETURN instruction
87
Figure 3.5: PC and Stack
3.7 SPECIAL FUNCTION REGISTER

The special function register are registers used by CPU and peripheral function for controlling the desired operation of the
devices. These registers are static RAM
Table 3.1: SFR Map
3.8 STATUS REGISTER

The STATUS register contains the arithmetic status of the ALU, the RESET status. The STATUS register can be the
destination for any instruction, like any other register. If the STATUS register is the destination for an instruction that affects
the Z, or C bits, then the write to these two bits is disabled. These bits are set or cleared according to the logic. Therefore, the
result of an instruction with the STATUS register as destination may be different than intended.
For example, CLRF STATUS will clear the upper-three bits and set the Z bit. This leaves the status register as
(where - = not implemented ).
Figure 3.6: Status Register

3.9 INDIRECT ADDRESSING:
INDF REGISTERS AND FSR REGISTERS.
The INDF register is not a physical register. Addressing the INDF register will cause indirect addressing. Indirect addressing
is possible by using INDF register. Any instruction using the INDF register actually access data pointed to by the file selected
88
register(FSR). Reading INDF itself indirectly will produce 00H. Writing to the INDF register indirectly result in no
operation(although status bit may be affected).
Movlw 0x0010;Initialize pointer
Movwf fsr ;to ram
Next:
Clrf indf ; Clear location
Incf fsr; incre pointer
Btfsc fsr, 4; All done?
Goto next; No clear next location
3.10 I/O PORTS

As with any register, the I/O registers can be written and read under program control. However, read Instructions (e. g.,
MOVF PORTB, W) always read the I/O pins independent of the pin’s input/output modes. On RESET, all I/O ports are
defined as input (outputs are at hi-impedance ) as the I/O control registers (TRISA,TRISB ) are all set.
PORTA And Data Direction Register (TRISA)
PORTA IS BIDIRECTIONAL. Port A is a 16 bit IO register. The output driver control register are loaded with the contents
of the W register by executing the TRIS f instruction. A ‘1’ from a TRIS register bit puts the corresponding output driver in a
high impedance mode. A ‘0’ puts the contents of the output data latch on the selected pins. A read of the ports reads the pins,
that is if an output driver on pin is enable and driver is high, but external system is holding it low, a read of the port will
indicate that the pin is low. The TRIS register are “write-only” and set to all ‘1’s (output driver disabled) upon RESET.[3]
Portb And Data Direction Register (Trisb)
PORTB IS BIDIRECTIONAL. Port B is a 16 bit I/O register. The output driver control register are loaded with the contents
of the W register by executing the TRIS f instruction. A ‘1’ from a TRIS register bit puts the corresponding output driver in a
high impendence mode. A ‘0’ puts the contents of the output data latch on the selected pins. A read of the ports reads the
pins, that is if an output driver on pin is enable and driver is high, but external system is holding it low, a read of the port will
indicate that the pin is low. The TRIS register are “write-only” and set to all ‘1’s (output driver disabled) upon RESET.
Bidirectional I/0 Ports

Some instruction Operates internally as a read followed by a write Operation. The BCF and BSF instructions,[3] for example,
read the entire port into the CPU, execute the bit Operation and write the result back to register. Caution must be used when
these instructions are applied to a PORT with both inputs and Outputs defined. For example, a BSF operation on bit 5 of
PORTB will cause all sixteen bits of PORTB to be read into the CPU. Then the BSF Operation takes place on bit 9 and the
PORTB valued is written to the output latches. if another bit PORTB is Used as a bi-directional I/0 pin (say bit 0)and it is
defined as an input at this time, the input signal present on the pin itself would be read into the CPU rewritten to the data
latch of this particular pin, overwriting the previous content. As long as the pin stays in the input mode, no problem occurs.
However, if bit0 is switched into output mode later on, the Content of the data latch may now be unknown.
; Initial PORT Setting
; PORT <15:8> Input
; PORT <7:0> Output
PORT latch PORT pins
BCF PORT, 7 ; pppp pppp 01pp pppp pppp pppp 11pp pppp
BCF PORT, 6 ;pppp pppp10pp pppp pppp pppp11pp pppp
BSF STATUS, R9 ;
MOVLW 03Fh ;
TRIS PORT B ;pppp pppp 10pp pppp
pppp pppp 10pp pppp
; Note that the user may have expected the pin value to be 00pp pppp . The 2nd ;BCF caused RB7 to be latch as the pin value
(High).
89
Figure 3.7: Flow chart of Read/Write port Data

4. DATA FLOW DIAGRAM
The data flow diagram is shown in Fig. 4.1. which shows the flow of data inside the Microcontroller while executing the
data transfer instructions. The data for other for other instructions flows in similar manner.[3]
Figure 4.1: The Data Flow diagram of Architecture
Entity block
Entity of the design defines its external interface and has following form[4]
ENTITY PIC_CORE IS
PORT (
clock, reset : in std_logic;
EA : in std_logic;
data_in_bus :inout std_logic_vector(15 downto 0);
Addr_out_pic :out std_logic_vector(15 downto 0);
PORTA : inout std_logic_vector(15 downto 0);
PORTB : inout std_logic_vector(15 downto 0)
)
End pic_core;
90
5. Experimental Result:
The program for the flow chart of Fig. 3.7 is designed to test the behavioral Model of RISC Microcontroller. Waveform in
the figure. 5.1 shows the (read and write)cycles of memory . Here PORTA and PORTB gives the ON/OFF control system.
Figure 5.1: Memory Read Cycle
6. CONCLUSION:
The Behavioral model of RISC Microcontroller is designed using VHDL. The Functional verification is done using Xilinx
tools[8]. The memory read/write cycle is shown in the figure 5.1. It is observed that the instruction executed in single cycle.
This is the attempt to design the 16 bit RISC Processor with additional instructions. Here few instructions are added in the
instruction set of PIC16c84.
7. REFERENCES
[1]. Kai Hwang “Advanced Computer Archicturer”, TMH Publication
[2]. PIC-18 ARCHITECTURAL OVERVIEW Available on htpp://www.microchip.com
[3] PIC 16/17 MICROCONTROLLER DATA BOOK MICROCHIP
[4] Perry, Douglas L. “VHDL”, Second edition McGraw-Hill,Inc., 1995.
[5] “The programmable Logic Data Book”, Xilinx Inc., San Jose, CA
[6]. “The programmable Gate Array Data Book”, Xilinx Inc., San Jose, CA
[7] “XACT User Guide”, Xilinx Inc., San Jose, CA
[8] “Foundation Series Quick Start Guide 1.5”, Xilinx Inc., San Jose, CA
[9] J. Gray, “FPGA CPU News”, www.fpgacpu.org/.
91
Two Phases Vaccination for Network Worms
Mehdi Goldoust Jildani1, Sam Jabbehdari2 and Amir Masoud Rahmani3

1
Islamic Azad University , North Tehran Branch, M_Goldoust@iau-tnb.ac.ir
2
Islamic Azad University, North Tehran Branch, S_Jabbehdari@iau-tnb.ac.ir
3
Islamic Azad University, Tehran Science and Research Branch, Rahmani@srbiau.ac.ir
Abstract
Network Worms are one of the most important and common Malware at Internet and Network that their spreading
velocity are very high considering to lack of direct interference of human. Automatic spreading fact causes to make
Infected most networks and computers by worms in a very short time and as a result the negative and unsuitable effect
of the worm in entire network will clear in a short time and cause to major damages to the entire network. Object of
presenting this paper is introduction of two phases vaccination for network worms that may be injected in entire
network and because of it in entire computers of the network and with consideration to type of its operation that is the
same renovation of vulnerable holes of machines, has made the network safe and through preventing the worms
spreading, it will make the network safe against these type of Malware. In order to study on effect of two phases
vaccination injection on network, a network that was connected with some vulnerable nodes to internet, was simulated
that the gained results showed high decrease of the numbers of the Infected nodes.
Keywords- Network worms, malware, vaccine, injection, Infected nodes, spreading, vulnerable.
92
1. Introduction
Network worm [19, 16, 14] is a self-repeated computer program [8] that has been produced on a computer and uses
computer networks and security holes present there in for their own reproduction [2]. Really, worm is a self-determined and
self-repeated pollution factor that has the ability for searching the new host for making it Infected [22, 21] using network [6].
A copy of worm has searched the network for finding another machine that has determined security hole and through finding
a vulnerable hole in new machine, will copy therein [3,5]. Through approaching the worm to a new machine, the cycle will
be repeated again consequently the worm is copied from this location at new machines, again, Really, the main difference
between worm and other destructive programs in being automatic of worm spreading [2].
The worms use different methods [7] in order to influence to offered machine and thereafter use them in order to
fulfillment their own codes that the following items may be mentioned:
1. Social engineering, encourages the email receivers for opening the enclosed files.
2. Weak figuring of the network; those networks that have open holes and ways in order to approach from out of
network.
3. Vulnerability in operating systems and softwares.
With consideration to different methods of worms spreading to the offered machine, vulnerability [20] at operating
systems and softwares is one of those factors that causes to spread the worms into the machines. Worms may spread after a
short time through profiting some systems having spread vulnerable holes [7] and their number will be increased every time
[10], in such a way that in a short-time will cause disorder at the operation of the network and consequently activity and
services of the network will encounter to disorder through spreading of numerous worms.
Producers of software after information from the vulnerable holes [10] at their products, through presenting required
patches, will take an action for removing problem and resolving the produced problem. All computer operators must be
confident from installation of the last presented patches related to software products that they use on their system. Such as
patches of a cloth that cause to improve the present holes at the software programs [10].
Software producers after preparing the patches, will present them on their websites [10, 11] and computer operators may
inform from submission the new patch through referring the site of supplier company of software product at the first step and
at the second step through its receiving and installation [10], will promote the installed software on their system. After
submission a patch, they must take an action to its installation on system, immediately and consequently, the opportunity to
use the vulnerable holes present at software products is deprived by the aggressor and decrease the possibility for their
success. Some of softwares are studies automatically required for submission the new and up to date copies and announce to
operators to present a new copy and there is possibility for its receiving and installation. Some of software producers will
inform the operators from preparation of a patch through email. If there will be possibility for use of the aforesaid facilities, it
is suggested to use its benefits. If there would not be the possibility for use of the mentioned potentials, producers websites
must be investigated periodically to inform from a new presented patch and take an action to receive and install it on their
systems, immediately.
2. Related Work
Most of the existing anti-virus techniques use a simple signature scanning approach to locate threats. As new viruses are
created, so do virus signatures. Smarter virus writers use more creative techniques to avoid detection. In response detection
mechanisms become ever more elaborate. This has led to coevolution, an ever-escalating arms race between virus writers and
anti-virus developers [17].
Recent incidents have demonstrated the ability of self-propagating code, also known as “network worms”, to infect large
numbers of hosts, exploiting vulnerabilities in the largely homogeneous deployed software base. Even when a worm carries
no malicious payload, the direct cost of recovering from the side effects of an infection epidemic can be tremendous. Thus,
countering worms has recently become the focus of increased research, generally focusing on content-filtering mechanisms
combined with large-scale coordination strategies [18, 17, 1].
Since the first Internet-wide worm, considerable effort has gone into preventing worms from exploiting common software
vulnerabilities by using the compiler to inject run-time safety checks into applications, safe languages and APIs, and static or
dynamic analysis tools. While shortcomings may be attributed to each of these tools or approaches individually, the fact is
that they have not seen wide use. We speculate that the most important reasons are: complexity; performance implications;
and, perhaps most importantly, a requirement for proactiveness on the part of application developers, who are often under
pressure to meet deadlines or have no incentive to use new-fangled software verification tools [18, 17].
Another approach has been that of containment of infected applications, exemplified by the “sandboxing” paradigm.
Unfortunately, even when such systems are successful in containing the virus, they do not always succeed in preventing
93
further propagation or ensuring continued service availability. Furthermore, there is often a significant performance overhead
associated with their use, which deters many users from taking advantage of them [18, 17].
3. Two Phases Vaccination
This vaccination is implemented in two phases:

1 - Vaccination against network known vulnerabilities.
2 - Vaccination against network unknown vulnerabilities.
Each of these phases is described respectively.
3.1 First Phase: Vaccination Against Network Known Vulnerabilities
In this phase, in order to better putting into operation the presented patches in order to renovation the holes and
Vulnerabilities of operating system or softwares that have been produced and presented by the operating system and software
producers. Operation of this phase is similar to operation of network worm and consists of fulfillment the following steps:
1. Injection of vaccine to the network.
2. Scan the network by vaccine in order to find computers having vulnerable holes.
3. Flowing the vaccine into computers having vulnerable holes through one of the received holes.
4. Scan and inspection each computer in order to find all vulnerable holes present therein.
5. Removing all worms in each computer by vaccine.
6. Applying and receiving the renovated patches related to each vulnerable hole from the server or website in order to
renovation the vulnerable holes.
7. Renovation the vulnerable holes through installation the patch related to each vulnerable hole.
8. Remaining of vaccine in each computer in order to contract with worms and renovation those vulnerabilities that
will be discovered in future.
3.1.1 Injection of vaccine to the network

In order to make active the vaccine of network worm, at first, the vaccine must be injected to the network. The vaccine
consists of all information related to all vulnerable holes, specification of all worms, all ways for spreading of each worm, all
ways related to renovation of each vulnerability, address of patches is specific for renovation of vulnerable that introduces its
information in each day using scan on internet sites that introduces the newest worms and makes security holes and patch of
each one up to date.
3.1.2 Scan of network by vaccine in order to find computers having vulnerable holes
With consideration to similarity of operation of this vaccine with worm, method of its scan [8] is similar to scan on worm
for finding Infected systems. Generally, there are different methods for finding the systems existing at the network.
As the worm must find an object in order to make it Infected, our vaccine in order to flow in each computer having
vulnerable hole [13], at first it must find it for seeking its vulnerable machine in the network. This vaccine such as worm,
uses one of the coincidental, statistical and topological scan and list of vulnerability in order to find the machines present at
the network.
3.1.3 Flowing the vaccine in the computers having vulnerable holes through one of the found holes
After finding the computers present at the network, it the turn of injection to the computer. For this reason, the considered
computer is scanned in order to find vulnerable holes. After finding the first vulnerable hole, the vaccine has used that hole in
order to be flowing the vaccine to it and consequently the vaccine will be injected to all computers having vulnerable hole.
3.1.4 Scan and inspection of each computer in order to find all vulnerable holes therein
After injection and flowing the vaccine in the computer having vulnerable hole, the vaccine take an action to scan on each
machine to determine all vulnerabilities in operating systems and softwares of each machine. The last known vulnerabilities
and specification of each one of vulnerabilities are introduced and made up to date by websites [10] every day[4].
94
3.1.5 Removing all worms in the computer by vaccine

With consideration to this point that in the previous step, all holes have been vulnerable, the vaccine has studied the
system, entirely to find the present worms. After finding the worms, it starts to remove the worms in total machine from these
worms that have used vulnerabilities of operating systems and softwares.
3.1.6 Applying and receive the renovating patches related to vulnerable hole from the server or website in
order to renovation the vulnerable holes
After discovery all vulnerable computers, it is turn of renovation the vulnerability hole [10] related to each vulnerability.
This vaccine in order to renovation the found holes, must receive at first the renovated patches of vulnerabilities from those
websites that introduce new vulnerabilities and also patches related to the vulnerabilities.
3.1.7 Renovation the vulnerable holes by installation of patches

After receiving the patches, it is the turn of installation of patches and consequently renovation the vulnerable hole. The
patches remove a specific problem or vulnerable point in a software and through installation of determined patches and
removing the vulnerable holes, possibility for spreading of those worms that use these holes have been removed and through
this method, it prevents spreading the worms into a machines and in a manner we cam claim that the network has been secure
against these types of worms.
3.1.8 Remaining of vaccine in each computer in order to contract with worms and renovation the
vulnerabilities that will be found in future
After removing all vulnerable holes and removing all worms from computer, Vaccine remains at the system and waits for
message from server in order to identification the vulnerability hole and new remained worm to renovate it.
This phase can step up to fix known vulnerabilities. To fix vulnerabilities unknown, the second phase of vaccination
should be carried out.
3.2 Second Phase: vaccination against network unknown vulnerabilities

The main goal of this phase is a vaccination against network unknown vulnerabilities. In this phase, in order to prevent
injuries and damage caused by vulnerabilities that have so far not been used ShieldGen [23] method. All cases have been
described in ShieldGen also is used in this system. The architecture introduced in ShieldGen can be concluded that the
system can reach up to very high vulnerability and identify unknown malwares with the exploitation of vulnerable into this
network to deal they published.
If not successful ShieldGen to identify unknown vulnerabilities, patches being produced from time to fix the vulnerability
by producers of softwares and operating systems, the first phase of vaccination using these patches to fix new vulnerabilities.
4. Simulation and Study Results

In order to simulation the vaccine operation and comparing the amounts of Infected nodes in those statuses that the worm
before injection of vaccine and after injection of worm is spread, ns 2.33 (NS2) Software has been used and in order to gain
the better results, internet and detailed network traffics have been experimented.
In ns [9], described a scalable worm propagationmodel, namely the detailed-network and abstract-network (DN-AN)
model. It combines packet-level simulations with analytic worm spreading model. The Internet with two parts: Detailed, and
Abstract part. A detailed-network could be an enterprise-network or the network run by an ISP. It simulates network
connectivity and packet transmission. Users can evaluate worm detection lgorithms in the detailed network. On the other
hand, we abstract the rest of the Internet with a mathematical model, namely susceptibleinfectious-removal (SIR) model
(refer to [15] for detailed descriptions). Compared to the detailed network, we only track several state variables in the abstract
world, such as the number of infected hosts. The interaction between DN and AN is through actual packet transmissions, that
is, the probing traffic generated by compromised hosts in both parts.
95
Table 1: Comparing the number of Infected nodes in the status of 50 nodes in

detailed network and 200000 nodes in abstract network
Time Infected nodes Infected nodes

(second) befor fist phase after fist phase
vaccination vaccination
55 0 0
56 0 0
57 0 0
58 0 0
59 1 0
60 3 0
61 5 0
62 8 0
63 12 0
64 13 1
65 18 2
66 22 3
67 25 4
68 26 5
69 28 6
70 30 7
In order to show the related results, this simulation was fulfilled in 70 seconds and in two specific statuses (before
injection of vaccine and after injection of vaccine) with 50 nodes (all of them are infected) in detailed network and 200000
nodes in abstract network (0.8 percent are infected) that numbers of Infected nodes have been shown before injection of
vaccine and number of Infected nodes after injection of vaccine in table 1.
Figure 1: Comparing the number of Infected nodes in status of 50 nodes in detailed network and
200000 nodes in abstract network
With consideration to the Figure 1, it is determined that injection of this vaccine has had its favorite effect on the network
and decreases the Infected nodes. If the simulation time will increase and more time will take to injection the vaccine, it will
have its effect in entire network and the network will be removed from these types of worms, entirely and the network will be
secure against them. ShieldGen simulation results [42] are indicative of positive performance.
96
5. Discussion
Following the description of our architecture, there are several other issues that need to be discussed.
The biggest challenge may be that this vaccine, like having a model release is a network worm and a worm network itself
be known. Because it Will be published into the computer without the user taking action to activity. To solve this problem,
the vaccine can do all their deeds ask permission from user.
Another challenge of this vaccine is to increase traffic on the network. Because the vaccine injection at any moment in the
entire network request to patch different websites that offer this patch them. This weakness can be covered with only one
time to download the patch and put it on a central server and can be programmed to receive all the Requested patches from
the server. In addition, vaccines can be planned and organized so that low-traffic time to do all their deeds.
Next topic, Which should get time to check to make patches to fix vulnerable holes of operating systems and software. For
increased performance and positive vaccine to prevent the exploitation of the hole to make vulnerable by network worms,
patches must be produced by manufacturers in the shortest possible time operating systems and software and are utilized for
repairing holes to make vulnerable.
Vaccine may be considered an antivirus or anti malware. But until that is installed antivirus or anti malware can to deal
with malware, and if being disabled antivirus or anti malware, malware without the slightest factor could come into
publishing networks. But once the vaccine being injected with the network to make and repair patches, even after being
inactive can help prevent getting into that kind of worms to use for publication by the security holes and weaknesses
operating system and software.
6. Conclusion
We have presented and introduced a vaccine in this article under title of " Two Phases Vaccination for Network Worms "
that has confronted against a type of network worms that influence and spread through weaknesses present at operating
systems and softwares into the network and prevent their arrival and spread into the machines. In this method, our
vaccination has used a set of factors and works in a special order that its result has been automatic renovation and patching of
all vulnerable holes of operating system and software used in machines to prevent the spreading of worms. Simulation of the
suggested plan indicates that the vaccine has been able to have a suitable operation in the network and secure the worms
spreading therein. Only a very basic point is that the patches must be produced at the shortest possible time and putting into
operation them in preventing the worm spreading. The next subject may introduce submission of some methods in order to
production and putting into operation the patches on time that may be a fact for an intelligent super vaccine.
References
[1] Stelios Sidiroglou, John Ioannidis, Angelos D. Keromytis, and Salvatore J. Stolfo. "An Email Worm Vaccine
Architecture" , To appear in the Proceedings of the 1st Information Security Practice and Experience Conference
(ISPEC) April 2005, Singapore
[2] Jiang Wu, Sarma Vangala, Lixin GaO, AND Kevin Kwiat. An E®ective Architecture and Algorithm for Detecting
Worms with Various Scan Techniques.
[3] THE RACE AGAINST MALICIOUS SOFTARE. SE 4C03 Winter 2005 Last Revised: April 2, 2005.
[4] http://www.cert.org/stats/vulnerability_remediation.html
[5] Zesheng Chen and Chuanyi Ji. Intelligent Worms: Searching for Preys.
[6] Jose Nazario, Defense and Detection Strategies against Internet Worms.2005
[7] Esharenana E. Adomi, Security and Software for Cybercafes, IGI Global snippet, 2008.
[8] Pankaj Kohli. Worms - survey and propagation. MS by Research - Computer Science and Engineering International
Institute of Information Technology Hyderabad, India.
[9] The VINT Project A Collaboration between researchers at UC Berkeley, LBL, USC/ISI, and Xerox PARC. The ns
Manual(formerly ns Notes and Documentation).January 6, 2009
[10] The Government of the Hong Kong Special Administrative Region. PATCH MANAGEMENT .February 2008
[11] Windows XP Service Pack 2.” (no date given), Microsoft Website. Retrieved March 27, 2005 from the World Wide
Web:http://www.microsoft.com/windowsxp/sp2/default.mspx
[12] Pankaj Kohli. Worms - survey and propagation.MS by Research - Computer Science and Engineering International
Institute of Information Technology Hyderabad, India.
[13] Craig Smith, Ashraf Matrawy,Stanley Chow and Bassem Abdelaziz. Computer Worms: Architectures, Evasion
Strategies, and Detection Mechanisms. Journal of Information Assurance and Security 4 (2009) 69-83.
[14] J. Shoch and J. Hupp. The “worm” programs – early experiments with a distributed computation. Communications of
97
the ACM, 22(3):172–180, March 1982.

[15] H. W. Hethcote. The mathematics of infectious diseases. SIAM Review, 42(4):599–653, October 2000.
[16] J. Brunner. The Shockwave Rider. Del Rey Books, Canada, 1975.
[17] Stelios Sidiroglou and Angelos D. Keromytis. "A Network Worm Vaccine Architecture" , In Proceedings of the IEEE
International Workshops on Enabling Technologies: Infrastructure for Collaborative Enterprises (WETICE), Workshop
on Enterprise Security. June 2003, Linz, Austria.
[18] Stelios Sidiroglou and Angelos D. Keromytis. "Countering Network Worms Through Automatic Patch Generation",
Columbia University technical report CUCS-029-03. November 2003, New York, NY.
[19] Siahn Qing and W. Wen, "A Survey and Trends on Internet Worms," Computers & Security, vol. 24, pp. 334-346, 2005
[20] Siahn Qing and W. Wen, "A Survey and Trends on Internet Worms," Computers & Security, vol. 24, pp. 334-346, 2005
[21] C. C. Zou,W. Gong, and D. Towsley. Code RedWorm Propagation Modeling and Analysis. In Proceedings of the 9th
ACM Conference on Computer and Communications Security (CCS), pages 138–147, November 2002.
[22] The Spread of the Sapphire/Slammer Worm. http://www.silicondefense.com/research/worms/slammer.php, February
2003.
[23] W. Cui, M. Peinado, H. J. Wang, and M. E. Locasto, "ShieldGen: Automatic Data Patch Generation for Unknown
Vulnerabilities with. Informed Probing," in 2007 IEEE Symposium on Security and. Privacy (SP '07), 2007
98
KMSVM model For Network Intrusion Detection
Arvind Mewada1 Sanyam Shukla2 Mansi Gyanchandani3

Department of Computer Science & Engineering
Maulana Azad National Institute of Technology
Bhopal (M.P.), India
1
mewadatec@yahoo.co.in
2
sanyamshukla@manit.ac.in
3
manasi_gyanchandani@yahoo.com
Abstract
Intrusion detection is a topic of interest in current scenario. Statistical IDS overcomes many pitfalls present in signature
based IDS. Statistical IDS uses models such as NB, C4.5 etc for classification to detect Intrusions. We have praposed KMSVM
model to perform multiclass classification. In this approach K-mean algorithm is used for selection of SVs to reduce the
learning time of MSVM. KDDCUP99 dataset is used for the experiment of intrusion detection and implementation of kernel
function for MSVM (1-versus-1, 1-versus-many and Error Correcting Output Coding (ECOC)) to perform classification for
statistical NBIDS. This paper explores the performance of MSVM for various categories of attacks.
Keywords: NBIDS; KDDCUP99; MSVM; Intrusion Detection; Kernel Function; RBF.
99
1. Introduction
This Intrusion detection (ID) is the processing procedure of identification and response to the action of malicious use
computers and network resources [1]. Intrusion Detection Systems (IDS) are computer programs that tries to perform intrusion
detection by comparing observable behavior against suspicious patterns, preferably in real-time. Intrusion is primarily network
based activity [2]. The primary aim of Intrusion Detection Systems (IDS) is to protect the availability, confidentiality and
integrity of critical networked information systems.
2. Techniques of IDS
2.1 Host-Based IDS and Network Based IDS
IDS can be classified based on which events they monitor, how they collect information and how they deduce from the
information that an intrusion has occurred. IDSs that operates on a single workstation are known as host intrusion detection
system (HIDS), A HBIDS adds a targeted layer to security to particularly vulnerable or essential systems, it monitors audit
trails and system logs for suspicious behaviors [3] while A network-based IDS monitors network traffic for particular network
segments or devices and analyzes network, transport, and application protocols to identify suspicious activity.
2.2 Misuse and Anomaly detection Techniques
Misuse detection uses the “signatures” of known attacks to identify a matched activity as an attack instance. Misuse
detection has low false positive rate, but unable to detect novel attacks. It is more accurate but it lacks the ability to identify the
presence of intrusions that do not fit a pre-defined signature, resulting not adaptive [4]. Misuse detection discovers attacks
based on patterns extracted from known intrusions [5]. Statistical based IDS: Statistical detection techniques assume that all
intrusive activities are necessarily anomalous. This means that if we could establish a "normal activity profile" for a system, we
could, in theory, flag all system states varying from the established profile by statistically significant amounts as intrusion
attempts. Anomaly detection is based on modeling the normal activity of the computer system. Unfortunately, the acquisition
of profiles of normal activity is not an easy task. The audit records used to produce the profiles of normal activity may contain
traces of intrusions leading to misdetections, and also activities of legitimate users often deviate from their normal profile as
modeled, leading to high false alarm rates [6].
2.3 Network Attack in IDS[7]
Denial of service[20] (DOS): In this type of attack an attacker makes some computing or memory resources too busy or too
full to handle legitimate requests, or denies legitimate users access to a machine. Examples are Apache2, Back, Land,
Mail bomb, SYN Flood, Ping of death, Process table, Smurf, Teardrop.
Remote to user (R2L): In this type of attack an attacker who does not have an account on a remote machine sends packets
to that machine over a network and exploits some vulnerability to gain local access as a user of that machine.
Examples are Dictionary, Ftp_write, Guest, Imap, Named, Phf, Send mail, Xlock.
User to root (U2R): In this type of attacks an attacker starts out with access to a normal user account on the system and is
able to exploit system vulnerabilities to gain root access to the system. Examples are Eject, Loadmodule, Ps, Xterm,
Perl, and Fdformat.
Probing: In this type of attacks an attacker scans a network of computers to gather information or find known
vulnerabilities. An attacker with a map of machines and services that are available on a network can use this
information to look for exploits. Examples are Ipsweep, Mscan, Saint, Satan, and Nmap.
3. K-Mean And MSVM
3.1 Basic of SVM
The theory of Support Vector Machine (SVM) is from statistics which is proposed by Vapnik. The basic principle of
SVM is finding the optimal linear hyperplane in the feature space that maximally separates the two target classes. For linearly
separable and non-separable data, it can be translated into quadratic programming (QP) and can get an only limit point. In the
case of non-linear, SVM can map the input to a high-dimensional feature space by using non-linear mapping and then the
linear hyperplane can be found [8].
3.2 SVM classification model
The basic principle of SVM is finding the optimal linear hyperplane in the feature space that maximally separates the
two target classes. The hyperplane which separates the two classes can be defined as:
100
Here xk is a group of samples:
, and k is the number of styles; n is the input dimension; w and b are

nonzero constants [9] [10].
Figure1. The optimal linear hyperplane (SV=Support vector)
3.3 Linearly separable model

Assume a training set:
, k is the number of samples. Thus, the problem can be described as:
Minimize (1)
Subject to .This is a quadratic programming (QP) problem. To solve it, we have to

introduce Lagrangian:
According to the Kuhn-Tucher conditions, we obtain
With the Lagrange multiplier for all i =1, 2… k. So the dual of equation (1) is:
For this problem, we also have the complement condition

.
So the optimal separating hyperplane is the following indicator function:
101
We can obtain the value of vector ω from (3).

3.4 Non-linear separable model
In the non-linear problem, it can be solved by extending the original set of variables x in a high dimensional feature
space with the map Φ. suppose that input vector x Є Rd is transformed to feature vector Φ (x) by a map Φ: Rd→H, then we
can find a function K (R’, R’) → R that satisfies condition K (xi, xj) = Φ(xi).Φ(xj) , so we can replace the inner-product
between two vectors (xi , xj )by K (xi, xj) and the QP problem expressed by (4) becomes:
The optimal separating hyperplane (5) can be rewritten as:
3.5 Multiclass support vector machine on non-linear separable model

Support vector machines are formulated for two class problems. But because support vector machines employ direct
decision functions, an extension to multiclass problems is not straightforward. There are roughly four types of support vector
machines that handle multiclass problems. But we use here only three for our research work (1-vs-many), pair wise coupling
(1-vs-1), and error-correcting output coding (ECOC) [11].
 One per class (OPC) also known as “one against others.” OPC trains K binary classifiers, each of which separates one
class from the other (K - 1) classes. Given a point X to classify, the binary classifier with the largest output determines
the class of X.
 The Pair wise coupling (PWC) constructs K (K-1)/2 pair wise binary classifiers. The classifying decision is made by
aggregating the outputs of the pairwise classifiers
 Error-correcting output coding (ECOC) [19] [12] used to reduce classification error by exploiting the redundancy of
the coding scheme. ECOC employs a set of binary classifiers assigned with codeword’s such that the Hamming
distance between each pair is far enough apart to enable good error correction.
3.6 K-means as a classical clustering algorithm

K-means is a classical clustering algorithm in the field of machine learning. It preserves the underlying structure and
distribution of the original data. K-means clustering technique is shown in following figure. There are twenty positive and
negative samples in figure, and they are compressed by the K-means clustering technique to only six cluster centers C1, C2…
C6. The statistical distribution and structure of the original data are preserved when represented by these six cluster centers.
This implies that there may be a lot of redundant information in the original data set, and it is possible to build classifiers with
acceptable testing accuracy based on those cluster centers compressed by the K-means clustering technique. This paper
combines the K-means clustering technique with SVM to build classifiers, and the proposed algorithm is called KMSVM. It
is possible for the KMSVM algorithm to build classifiers with many fewer support vectors and higher response speed than
SVM classifiers.
102
Figure2.K-means clustering is run on the 2-dimensional data set and six clusters are formed, C1,C2…, C6. H is the classifier
separating two-class samples into Class1 and Class2
4. Proposed Approach
To separate attacks from legitimate activities, all of the machine learning based intrusion detection technologies will have two
main phases, training procedure and detection procedure. As shown in Fig. 3, in the training procedure of KMSVM, K-mean is
used to extract the optimal discriminate support vectors of the whole training data. In MSVM for making decision function
needs support vectors other vectors far from decision boundary useless for MSVM.
Figure3. Architechure of KMSVM

K-mean clustering algorithm is used to cluster the projected training datasets and remove unused data by using the
information provided by the K-mean clustering results, to setup the multiclass SVM detection model. The detection model
consists of many binary detection models. Each binary detection model includes two main types of information, the optimal
discriminate support vectors extracted from the sub training data by using K-mean and the MSVM classifier based on the
projected sub training data. In the detecting procedure, we project the test dataset according to the detection model and then
classify the data as normal or malicious by using the detection model.
5. Dataset And Experiments

The KDD Cup 1999 uses a version of the data on which the 1998 DARPA Intrusion Detection Evaluation Program was
performed. Each instance in the KDD Cup 1999 datasets contains 41 features that describe a connection. Features 1-9 stands
for the basic features of a packet, 10-22 for content features, 23-31 for traffic features and 32-41 for host based features.
There are 38 different attack types in training and test data together and these attack types fall into five classes: normal,
probe, denial of service (DoS), remote to local (R2L) and user to root (U2R) [14].
103
TABLE I. DATASET
Dataset Train Records Test Records
KDD CUP99 48,984,31(4.9 million) 3,11,029(0.3 million)
In this experiment we use Pentium (IV 3GH) processor, 512 MB RAM, running window XP (SP2) based multiclass
SVMlight [15]. We have implemented Cauchy and ANOVA kernel functions. For Cauchy and ANOVA kernels, the
accuracy for all the three MSVM methods was very low. We exclude the results for these. The experiment using
RBF[16][17][18] kernel function for intrusion detection (multiclass classification) with parameters as g=0.001, c=1, q=50,
n=40 gave the intrusion detection rate 92.05%, 90.65% & 92.00% for one-vs.-one, one-vs.-many & ECOC MSVM methods
respectively.
6. Evaluation Matrics
The Evaluation Metrics mainly used following steps to evaluate the performance of classifier:
 . True positives: The true positives (TP) and true negatives (TN) are correct classifications.
 False positive: A false positive (FP) occurs when the outcome is incorrectly predicted as yes (or positive) when it is
actually no (negative).
 False negative: A false negative (FN) occurs when the outcome is incorrectly predicted as negative when it is actually
positive.
 Recall: The percentage of the total relevant documents in a database retrieved by search. If user knew that there were
1000 relevant documents in a database and his search retrieved 100 of these relevant documents, his recall would be
10%.
Recall =TP/ (TP+FN)
 Precision: The percentage of relevant documents in relation to the number of documents retrieved. If search retrieves
100 documents and 20 of these are relevant, then precision is 20%.
Precision=TP / (TP+FP)
 F-measure: The harmonic mean of precision and recall
F = 2 * Recall * Precision / (Recall + Precision)
The true positive rate is TP divided by the total number of positives, which are TP + FN. The false positive rate is FP
divided by the total number of negatives, FP + TN.
ROC area in ROC analysis we plot true positive ratio (tpr) against, false positive ratio (fpr).The overall success rate is the
number of correct classifications divided by the total number of classifications:
In a multiclass prediction, the result on a test set is often displayed as a two dimensional confusion matrix with a row and
column for each class. Each metrics element shows the number of test examples for which the actual class is the row and the
predicted class is the column. Good results correspond to large numbers down the main diagonal and small, ideally zero, off-
diagonal elements
TABLE II. CONFUSION MATRICS

Predicted Class
Yes No
Yes True Positive False Negative
Actual Class
No False Positive True Negative
104
7. Conclusion and Future Work
There are many kernel functions which can be used for intrusion detection purpose. Among those we have implemented
Cauchy and ANOVA kernel functions. We have conducted experiment using Cauchy, ANOVA and RBF kernel function
over three types of MSVM. And found that the RBF kernel function’s performance is better for intrusion detection. The
intrusion detection rate is 92.05%, 90.65% & 92.00% for one-vs.-one, one-vs.-many & ECOC methods respectively using
RBF kernel function. This result can further be improved by using composite (combine two kernel function) of two or more
kernel functions.
References
[1] Jai Sunder balasubamaniyan, Jose Dmar Garcia-Fernandez, David Isacoffet.al, “An Architecture for Intrusion Detection
using Autonomous Agents,” COAST Laboratory, Purdue University, COAST Technical Report june 1998.
[2] S. Axelsson, “Research in intrusion-detection systems: A survey,” Department of Computer Engineering, in Chalmers
University of Technology, December15, 1998.
[3] Y. Liu, D. Tian, and B. Li, 2006, "A Wireless Intrusion Detection Method Based on Dynamic Growing Neural
Network," presented at Proceedings of the First International Multi-Symposiums on Computer and Computational
Sciences (IMSCCS'06).
[4] K. Ilgun, R. A. Kemmerer, and P. A. Porras, March 1995,“State transition analysis: A rule-based intrusion detection
approach,” IEEE Transactions on Software Engineering, 21(3):181–199.
[5] D. Barbara and S. Jajodia, 2002,“Applications of Data Mining in Computer Security,” Norwell, MA: Kluwer.
[6] G. Giacinto and F. Roli, 2002, “Intrusion detection in computer Networks by multiple classifier systems”, in Proc. Int
Conf. Pattern Recognition.
[7] M. Peddabachigaria, Ajith Abrahamb, Crina Grosanc, Johnson Thomasa 8 June 2005,“Modeling intrusion detection
system using hybrid intelligent systems”, Elsevier and Science Direct , Journal of Network and Computer Applications.
[8] Hui Zhao, Yong Yao, and Zhijing Liu Liu, 2007, “A Classification Method Based on Non-linear SVM Decision Tree”,
Fourth International Conference on Fuzzy Systems and Knowledge Discovery.
[9] Kuan-Ming Lin and Chih-Jen Lin, NOV 2003, “A Study on Reduced Support Vector Machines”, IEEE Transactions On
Neural Networks, VOL. 14, NO. 6.
[10] R.Debnath, H.Takahashi, July 16, 2006, “SVM Training: Second-Order Cone Programming versus Quadratic
programming”, IEEE International Joint Conference on Neural Networks, Canada, pp.1162-1168.
[11] K. Goh, E. Chang, K. Cheng, November 10, 2001,”SVM Binary Classifier Ensembles for Image Classification,”
Atlanta, Georgia, USA,CIKM’01.
[12] Bernhard Pfahringer, January 2000, “Winning the KDD99 Classification Cup: Bagged Boosting”, ACM SIGKDD
Explorations Newsletter, Volume 1, Issue 2, p. 65-66.
[13] “http://kdd.ics.uci.edu/databases/kddcup99/task.html”, KDD Cup1999.
[14] T. G. Dietterich and G. Bakiri, 1995, “Solving multiclass learning problems via error-correcting output codes”, Journal
of Artificial Intelligence Research, 2:263-286.
[15] Thorsten Joachims Cornell University, Department of Computer Science, “http://svmlight.joachims.org”.
[16] M. Bianchini, P. Frasconi, and M. Gori, May 1995, “Learning without local minima in radial basis function networks,”
IEEE Tranaction. Neural Network., vol 6, no. 3, pp. 749–756.
[17] C. M. Bishop, 1991 ,“Improving the generalization properties of radial basis function neural networks,” Neural
Computat., vol. 3, no. 4, pp. 579–588,.
[18] M. J. L. Orr, 1996, “Introduction to radial basis function networks,” Center Cognitive Sci., Univ. Edinburgh, Edinburgh,
U.K.
[19] Andrea Passerini, Massimiliano Pontil, and Paolo Frasconi, Member, IEEE, January 2004,”New Results on Error
Correcting Output Codes of Kernel Machines” IEEE Transactions on Neural Networks, Vol. 15.
[20] Srinivas Mukkamala, Andrew H. Sunga, Ajith Abraham, 7 January 2004, ”Intrusion detection using an ensemble of
intelligent paradigms”, Elsevier, Science Direct.
[21] Jiaqi Wang, Xindong Wu, Chengqi Zhang, 2005, “Support vector machines based on K-means clustering for real-time
business intelligence systems”, Int. J. Business Intelligence and Data Mining, Vol. 1, No. 1.
105
Improving Security and Reducing Computation Time using the ECDSA and PHAL
Ranjeet Osari1, Prof. Bhola Nath Roy2
CSE Department, Maulana Azad National Institute of Technology ,Bhopal (M.P.)1,2
ranjeet.osari@gmail.com1
bhola_mact@rediffmail.com2
Abstract
The internet has been spread in all areas for the information transmitting and accessing for widely uses, The organizations
increase their confidence on, possibly circulated, information systems for daily business, they become more vulnerable to
security breaches even as they gain productivity and efficiency advantages. it becomes big issues that how to protect the
confidential data or information while transmitting and receiving. The security is based on the reliability, functionality and
accuracy of the system. Most of the security system are using the RSA digital signature for the authentication and many
encryption decryption methods for secret sharing the data or information on the internet. In this paper the advance
technology ECDSA and PHAL used for the higher security system to protect the data from the unauthorized users and reduce
the computation time.
Keywords: RSA, Hash function, ECDSA, PHAL, Digital Time Stamping.
106
1. Introduction
The complete secure system is based on the security goals confidentiality, integrity and authentication. These goals can
be completed by using the digital signature encryption and decryption methods with time stamping. Though a number of
methods, are currently available to protect data while transmitting across the network, These methods provide the security but
it consume the more time .for reducing the computation time we are using the ECDSA and using the PHAL for more security
and it provide the better result with compare to the other secure system. The digital signature scheme based on elliptic curve
and proactive secret sharing. This scheme make up for the lack of exiting schemes, possessed stronger security, and from the
view of application, the new scheme was more practical. And the PHAL is more delicate hash algorithm to compare with the
SHA/MD. PHAL reduce the weakness of the SHA/MD and provide the high security.
2. Digital signatures scheme

A digital signature is a mathematical scheme applying for the authenticity of a electronic data or document the digital
signature also called the electronic signature . In the digital signature it perform an asymmetric key operation where the
private key is used to digitally sign an electronic document and the public key is used to verify the signature. Digital
signatures provide authentication and integrity protection.
There are main three Digital signature RSA,DSA and ECDSA.
2.1 RSA
The RSA (Rivest-Shamir-Adleman) public-key cryptosystem involves exponentiation
modulo a number n that is the product of two large prime numbers. Plaintext is encrypted in blocks, with each block having a
binary value less than the number n. That is, the block size must be less than or equal to log2(n); in practice, the block size is
i bits, where 2i<n≤2i+1.Encryption and decryption are of the following form, for some plaintext block M and cipher text
block C:
C = M e mod n
M = C d mod n = (M e) d mod n = M ed mod n
Both sender and receiver must know the value of n. The sender knows the value of e, and only the receiver knows the value
of d. Thus, this is a public-key encryption algorithm with a public key of PU = {e,n} and a private key of PR ={d,n}.When
referring to the key size for RSA, what is meant is the length of the modulus n in bits. A typical key size for RSA is 1024
bits. RSA can be used for encryption and also be used for digital signature[1].
2.2 DSA
DSA (Digital Signature Algorithm) suitable for applications requiring a digital rather than hand written signature. the
DSA is public-key techniques based on exponentiation modulo a large prime number p. For this scheme, the key size is the
length of the prime p in bits, and a typical value is 1024 bits. The digital signature is computed using a set of rules and a set
of parameters such that the identity of the participant and integrity of the data can be verified. The DSA provides the
capability to generate and verify signatures. The digital Signature generated by the help of private key and the Signature
verification take place with the help of public key, Each user use a private key and public key pair. Public keys are implicit
to be known to the public in general. Private keys are never shared. Anyone can verify the signature of a user by employing
that user's public key. Signature generation can be performed only by the use of the user's private key.
the specifications for testing the individual DSA components of the IUT. These components are Domain Parameter
Generation, Domain Parameter Verification, Key Pair Generation, Signature Generation, and Signature Verification[2]
2.3 ECDSA
The Elliptic Curve Digital Signature Algorithm (ECDSA) An ECDSA key pair is associated with a particular set of EC
domain parameters. the elliptic curve cryptosystem basic principle, and then proposes a password authentication scheme
based on the elliptic curve cryptosystem. User’s password is utilized ECC to encrypt in the process of storage and
transmission. According to the difficulty of elliptic curve discrete logarithm, it is hard to explain the password.
Authentication information keeps dynamic, so replay attack can be prevented effectively[3]
The public key is a random multiple of the base point, while the private key is the integer used to generate the multiple.
Key Pair Generation :- An entity A’s key pair is associated with a particular set of EC domain parameters
D=(q.FR,a,b,G,n,h). each entity A does the following.
 Select a random or pseudorandom integer d in the interval [1,n-1].
 Compute Q=dG.
 A’s public key is Q; A’s private key is d.
107
Public Key Validation:- Public key validation, as first enunciated by Johnson , ensures that a public key has the requisite
arithmetical properties. Successful execution of this routine demonstrates that an associated private key logically exists,
although it does not demonstrate that someone actually has computed the private key nor that the claimed owner actually
possesses the private key.
The assurance that a public key Q is valid can be provided to an entity A using one of the following methods:
 A performs explicit public key validation using Algorithm (shown below).
 A generates Q hit self using a trusted system.
 A receives assurance from a trusted party T (e.g. a Certification Authority) that T has performed explicit public key
validation of A using Algorithm.
 A receives assurance from a trusted party T ã that Q was generated using a trusted system.
Algorithm: - Explicit validation of an ECDSA public key.
INPUT: A public key Q=(XQ.YQ) associated with valid domain parameters (q.FR,a,b,G,n,h).
OUTPUT: Acceptance or rejection of the validity of Q.
Check that Q≠O.
Check that XQ and YQ are properly represented elements of Fq (i.e., integers in the interval [0,p-1] in the case q=p, and bit
strings of length m bits in the case q=2m).
Check that Q lies on the elliptic curve defined by a and b.
Check that nQ=O.
If any check fails, then Q his invalid ; otherwise Q his valid[4].
3. Hashing Algorithms
The Hashing functions are used to compress an random length message to a fixed size, usually for following signature by
a digital signature algorithm. hash function h should have the following property:
 h should destroy all homomorphism structures in the underlying public key cryptosystem .
 h should be computed on the entire message
 3h should be a one-way function so that messages are not disclosed by their signatures.
 it should be computationally infeasible given a message and its hash value to compute another message with the same
hash value.
 should resist birthday attacks (finding any 2 messages with the same hash value, perhaps by iterating through minor
permutations of 2 messages)
3.1 MD2, MD4 and MD5

The family of one-way hash functions by Ronald Rivest. MD2 is the oldest, produces a 128-bit hash value, and is
regarded as slower and less secure than MD4 and MD5.
MD4 produces a 128-bit hash of the message, using bit operations on 32-bit operands for fast implementation.
MD4 overview
 pad message so its length is 448 mod 512.
 append a 64-bit message length value to message.
 initialize the 4-word (128-bit) buffer (A,B,C,D)
 process the message in 16-word (512-bit) chunks, using 3 rounds of 16 bit operations each on the chunk & buffer.
 output hash value is the final buffer value.
Some progress at crypt analysing MD4 has been made, with a small number of collisions having been found. MD5 was
designed as a strengthened version, using four rounds, a little more complex than in MD4. a little progress at crypt
analysing MD5 has been made with a small number of collisions having been found. both MD4 and MD5 are still in use
and considered secure in most practical applications. both are specified as Internet standards (MD4 in RFC1320, MD5 in
RFC1321)[5].
3.2 SHA (Secure Hash Algorithm)

SHA is a close relative of MD5, sharing much common design, but each having differences.SHA has very recently been
subject to modification following NIST identification of some concerns, the exact nature of which is not public, current
version is regarded as secure. it produces 160-bit hash values.
108
SHA overview
 pad message so its length is a multiple of 512 bits
 initialize the 5-word (160-bit) buffer (A,B,C,D,E) to (67452301,efcdab89,98badcfe,10325476, c3d2e1f)
 process the message in 16-word (512-bit) chunks, using 4 rounds of 20 bit operations each on the chunk & buffer.
 output hash value is the final buffer value.
3.3 PHAL
Family of parameterized hash algorithms - PHAL is a proposal of a new dedicated hash algorithm designed as an answer
to weaknesses of MD/SHA hash function family. Recently proposed attacks on well known and widely used hash functions
motivate a design of new hash functions. where few elements of hash function are parameterized. This approach makes the
hash algorithm more secure and more flexible. PHAL consists of two mechanisms: new iteration schema and dedicated
compression function.[6]
For PHAL hash algorithm a similar approach was chosen. Additionally, the number of rounds was added as a parameter. The
design goals of this hash algorithm are determined as follows:
 Hash algorithm must provide message digests of 224, 256, 384 and 512 bits and shall support a maximum message length
of at least 264–1 bits [7].
 Its iteration structure should be resistant against known attack against the MD-type structure.
 Its compression function should be resistant against known attack.
 Its structure should be parameterized, to reach flexibility between performance and security.
4. Digital time stamping

The time stamping is a method through the user can calculate the time that how much time the user take to complete the
process. The digital time stamping show the digital document when created and modified by the user, it certified the
particular record at the particular time.
Two features of a digital time-stamping system are particularly helpful in enhancing the integrity of a digital signature
system. First, a time-stamping system cannot be compromised by the disclosure of a key. This is because digital time-
stamping systems do not rely on keys, or any other secret information. Second, digital time-stamping certificates can be
renewed so as to remain valid indefinitely [8].
5. The Enhance Scheme of Digital Signature

The enhance scheme of digital signature used in this paper for providing the high security in security model while using
the ECDSA. Elliptic curves are an important branch of algebra and geometry. The research on elliptic curves has been taking
on for many years and has been obtained plentiful productions.
Cryptography has a great interesting in elliptic curves(E) in finite fields(F), such as: F(p), F(2m), F(pm). Finite fields are
identified with the notation F(pm), where p is a prime and m is a positive integer. It is well known that finite fields exist for
any choice of prime p and integer m. The addition of points on an elliptic curve is defined with chord contact method, which
made Abel groups. Elliptic curve cryptosystems (ECCs) is based on the Abel group[9]
109
Figure 1: The Security Model

Considering these factors above, I propose a scheme of digital signature which based on the combination of symmetric and
asymmetric cryptography. Specifically, 3DES will be selected for encrypting sources, and ECDSA will be selected for digital
signature and transmitting secret key (the key for symmetric cryptography is called secret key and the keys for asymmetric
cryptography are called public key and private key). Of course hash function and digital time-stamp must also be applied in
this scheme. Fig.1 displays this scheme Something should be pay attention to: (1) This scheme is a simplified model without
PKI (Public-Key Infrastructure), So I assumed that both sender and receiver have already obtained the public key from each
other. Respectively the public and private keys are (Ka1, Ka2) belong to the sender, (Kb1, Kb2) belong to the receiver. (2) In
the middle of the figure 1 is a heavy line that represents communication network; upward side is the sender; under side is the
receiver. The dotted curve represents transmitting process.
Steps in this scheme:
Step 1: Electronic document is processed by PHAL function that generates PHAL value. Then both PHAL value and digital
time-stamp will be signed by ECDSA by means of private key Ka1 belongs to the sender. At the same time, electronic
document is encrypted by 3DES that produces cipher text. Besides, secret key will be encrypted by ECDSA by means of
public key Kb2 belongs to the receiver, and it forms a cipher block (a small light grey rectangular in the right hand of Figure
1). Both digital signature and cipher text are transmitted from sender to receiver.
Step 2: The digital signature and cipher text are separated by receiver. Firstly, digital signature is
verifying by means of public key Ka2 belongs to the sender that produces hash value and digital time-stamp. Secondly,
PHAL value will be signed by ECDSA by means of private key Kb1 belongs to the receiver whose action is doing reply to
the sender.
110
Step 3: After obtaining the digital signature form receiver (a small dark grey rectangular in the left hand of Figure 1), which
will be verifying by means of public key Kb2 belongs to the receiver. Then the sender can obtain a PHAL value which must
be compared with the step 1’s PHAL value. The cipher block in step 1 will be transmitted to receiver only if the two PHAL
values are identical with each other.
Step 4: The cipher block decrypted by ECDSA by means of private key Kb1 from the receiver. Then the receiver can obtain
the secret key which can be used for 3DES decryption for the cipher text. As a result, electronic document can be obtained.
Step 5: The electronic document from step 4 should be processed by PHAL function that produces PHAL value. Compare
this PHAL value with step 2’s PHAL value, at the same time check to the digital time-stamp obtained from step 2. The
receiver can accept the electronic document only if both PHAL values are identical and digital time-stamp is accurate [1].
6. Conclusion
In this research founded the ECDSA is highly secure then other Standard Digital signature scheme and the PHAL
algorithm reduce the time for computation. In this research proposed the scheme improve the security and reducing the
computation time. This scheme is an enhanced digital signature scheme that provides the high security and it reduces the
computation time and increases the speed of the security models.
References
[1] Na Zhu, GuoXi Xiao (2008) “The Application of a Scheme of Digital Signature in Electronic Government”
International Conference on Computer Science and Software Engineering.
[2] Lawrence E. Bassham III (March 10, 2004) “The Digital Signature Algorithm Validation System (DSAVS)”
National Institute of Standards and Technology Information Technology Laboratory Computer Security Division
[3] Wei Haiping, Jia Chuanying (2007) “The Study of Password Authentication System Based on Elliptic Curve
Cryptosystem” Navigation college, Dalian Maritime University Dalian,Liaoning 116026,China Liaoning Shihua
University Fushun,Liaoning 113001,China IEEE.
[4] Don Johnson and Alfred Menezes and Scott Vanstone “The Elliptic Curve Digital Signature Algorithm (ECDSA)”
Certicom Research, Canada Dept. of Combinatorics & Optimization, University of Waterloo, Canada certicom.
[5] R L Rivest, (1991) "The MD4 Message Digest Algorithm", Advances in Cryptology - Crypto'90, Lecture Notes in
Computer Science No 537, Springer- Verlag, pp303-311
[6] P. Rodwald, J. Stoklosa, (2008) "PHAL-256 ? Parameterized Hash Algorithm," ias, pp.50-55, 2008 The Fourth
International Conference on Information Assurance and Security,
[7] NIST, (2007) Announcing Request for Candidate Algorithm Nominations for a New Cryptographic Hash Algorithm
(SHA-3) Family. http://www.nist.gov/hashcompetition
[8] Stuart Haber, Burt Kaliski and Scott Stornetta, ( 1995) “How do digital timestamps support digital signatures?”
CryptoBytes, vol. 1, no. 3, RSA Laboratories, pp. 14-15.
[9] Ping Guo, Hong Zhang (2009) “An Elliptic Curve Digital Signature Scheme Based on Proactive Secret Sharing”
College of Computer and Software Nan Jing University of Info Science and Tech Nan Jing, China IEEE.
111
Risk Assessment for E-commerce Security based on Fuzzy Iteration Model

Rijvan Beg1, R. K. Pateriya2, S.C. Shrivastava3
1
M.Tech (Information Security) Scholar, Maulana Azad National Institute of Technology,
2
3
Professor, Electronics Engg. Department, Maulana Azad National Institute of Technology,
rijvan.beg@gmail.com, pateriyark@gmail.com, scs_manit@yahoo.co.in
Abstract
With the hasty development of e-commerce, security issues are arising from people’s attention. E-commerce services
have risen to become a core element of the internet and web environment. Exchange security on network is very important
for e-commerce services and it is always the key factor that affects the accomplishment of e-commerce. Risk analysis is
mainly performed using risk matrix method and the method based on CIM model. Most of risk factor in e-commerce is
uncertain and fuzzy and its risk is difficult to be calculated by classical mathematics. In this paper a novel risk analysis model
is proposed based on fuzzy iteration. The novel approach is quite practical in evaluating the risk on different security
parameter.
Keywords: e-commerce; security analysis; CIM model; Fuzzy logic, fuzzy iteration model.
112
1. Introduction
With the active expansion of internet technology, e-commerce came into being and prompts growth which
based on the network and multimedia technology. The e- commerce is through public network, such as internet,
open computer network to conduct online transactions, which can fast and effective to implement variety of
business process [1].
With the online business extension of e-commerce, e-commerce security becomes more and more
marvelous. How to create the safe and expedient e-commerce application environment and provide the strong
protection of information security has become a spotlight of e-commerce [1].
E-commerce is faced with various risks. It is important that you understand the risks facing your e-
commerce system, and the potential impact of any security incident.
The materialization of e-commerce may not only exaggerate the occurrence possibility and the dent of
various types of risks, but also bring new risks in the area of financial services. While observing to e-commerce
risk management, the western developed countries by research and practice, have accumulated rich experiences
and a lot of standardized norms, however, the research and practice of e-commerce risk management are
relatively backward [3].
In risk assessment, the most important factor is to decide what is a hazard in the workplace. If the risk is
determined to be significant enough, precautions should be put into place so that the risk is minimized or
dispensed with altogether.
Corporations face increased levels of risk almost daily: from software vulnerabilities hidden in their
business-technology systems to hackers and cyber crooks trying to steal proprietary corporate intellectual
property, including sensitive customer information. An ever-growing list of government regulations aimed to
ensure the confidentiality, integrity, and availability of many types of financial and health-related information
also is increasing IT risks and making a comprehensive security risk assessment a modern day corporate
necessity
2. Problem Faced in E-Commerce

Appropriate to the complexity and vulnerability of the internet, the development of e-commerce based on
internet is faced with serious security problems. Usually, there are several kinds of security risks as follows [5-
6].
1.1 Tempering with Data.
Data tampering is the deliberate destruction or manipulation of data. The tampering may or may not be
detected until some time in the future. Data is threatened both in-transit (physically or electronically) and while it
is stored.
There are many techniques to temper the data and means, network attackers mainly tamper with data and
demolish the internet software make the information system fail, and they also often copy, tamper with, delete or
insert transmitting information in order to undermine the integrity of information [1].
1.2 Data Access Risk.
Data access risk primarily is the result of both the data revision and deletion from the unauthorized access to
the database system and the operational mistakes of the staff in the e-commerce [1].
1.3 Fake Information.
A criminal activity which involves stealing some WEB PAGES and placing them on your site in order for
the site to be placed in a prominent position when a search using a search engine is carried out.
1.4 Online Payment Risk.
Protecting card details is the primary security risk with electronic transactions. Customers are very
comfortable using cards in shops and over the phone despite the ever-present risk of details being copied or
stolen. With payments over the Internet, there is more resistance towards disclosing card information.
113
Hackers can often easily dupe users accessing online payment systems and banking systems into revealing
their account information after receiving official-looking e-mails. But again, e-mail is pretty easy to forge. And
even an intelligent user may have difficulty discerning an official e-mail from a forgery.
3. Approach Used for Security Measurement and Risk Analysis.
3.1 Risk Matrix.
Risk matrix is the most straight and helpful way of risk analysis. The occurrence probability of risk and the
loss of assets are respectively regarded as the levels of rows columns, with the degree higher, high, medium, low,
and lower. The number at the cross of the matrix shows the risk.
Supposing that risk probability and loss degree are in the lower order from 10 to 0, respectively 10, 8,6,4,2,
then the risk is the product of two factors. Risk analysis matrix is shown in table 1.
Table 1: Risk Analysis Matrix
PROBA- HIGHER HIGH MEDIUM LOW LOWER

BILITY
LOSS
DEGREE
HIGHER 10,10 10,8 10,6 10,4 10,2
HIGH 8,10 8,8 8,6 8,4 8,2
MEDIUM 6,10 6,8 6,6 6,4 6,2
LOW 4,10 4,8 4,6 4,4 4,2
LOWER 2,10 2,8 2,6 2,4 2,2
3.2 Risk Analysis Based on CIM.

Fuzzy logic is a form of multi-valued logic derived from fuzzy set theory to deal with reasoning that is
approximate rather than precise. In contrast with "crisp logic", where binary sets have binary logic, the fuzzy
logic variables may have a membership value of not only 0 or 1 – that is, the degree of truth of a statement can
range between 0 and 1 and is not constrained to the two truth values of classic propositional logic.[1]
Furthermore, when linguistic variables are used, these degrees may be managed by specific functions.
Fuzzy logic emerged as a consequence of the 1965 proposal of fuzzy set theory by Lotfi Zadeh. Though fuzzy
logic has been applied to many fields, from control theory to artificial intelligence, it still remains controversial
among most statisticians, who prefer Bayesian logic, and some control engineers, who prefer traditional two-
valued logic.
3.2.1. Trapezoidal fuzzy number: In the domain X, the trapezoidal fuzzy number is as follows:
3.2.2. CIM Model: CIM model (control interval and memory model) is a kind of risk analysis model put
forward by American scholars coopor and chapman. There are two types: the serial response model and the
parallel response model, which were respectively the serial or parallel some of variable probability distribution
[1].
114
If an event has n risk factors and the occurrence of each factor will separately influence the event, then the
probability distribution model with n risk factors is called parallel response model [1].
3.3.3. E-commerce Risk Assessment of E-commerce Based on CIM mode: This is indicated by a function of
the probability of risk event and its impact, that is, R=f (P, I), where R shows risk, P the probability of risk event,
I the impact of risk incidents. In view of the characteristic of all risk factors in e-commerce of e-commerce, the
CIM model is applied to e-commerce risk assessment of e-commerce as follows:
a) Constructing risk factor sets and judgment sets: Risk factor sets and judgment sets of bank e-commerce
are constructed, and different judgment sets can be established for the occurrence probability of risk factors and
the impact of risk factors. Assumption that the risk factors set is Fi = (f1,f2,…….,fn), i = 1,2,3 and the judgment
set is J = (j1,j2,……,jm), the qualitative comments in the judgment set are expressed by trapezoid fuzzy number.
b) Quantifying the fuzzy evaluation of risk factors: The expert evaluation method is used to determine
fuzzy evaluation of the judgment set of the occurrence probability of risk factors and the impact of risk factors of
e-commerce. Fuzzy evaluation of risk factors by the fuzzy way is processed to get the probability distribution
range and the impact distribution range, at the same time calculate expectation of individual risk factor judging
the individual risk factor [1].
c) CIM Calculation: CIM model is used to calculate the probability distribution range and the impact
distribution range of all types of e-commerce risks, by which the expectation of the overall risk can be calculated
to assess the e-commerce risk [1].
4. E-Commerce risk management strategies

Because of the importance of e-commerce security, it is necessary for e-commerce to be provided with the
complete and effective e-commerce risk management strategy to forecast and control effectively the risk which
e-commerce probably are confronted. The strategies mainly include the following [1].
4.1 Physical Security.

Physical security is not in frequent use nowadays.. Physical security exists in order to deter persons from
entering a physical facility. Historical examples of physical security include city walls, moats, etc
It can be as simple as a locked door or as elaborate as multiple layers of armed Security guards and
Guardhouse placement
Still there are requirements of physical security for the security of e-commerce systems. It can protect
computer system, e-commerce server, and communication media and so on from the various risks.
4.2 Virus Surveillance.

Virus surveillance is a mechanism that find the viruses in side the system and protect the system, such as the
firewall technology which provide the security from the viruses, control and identify the access to a variety of
sites.
4.3 Data Defence.

Data defence consist of the techniques which reduces the data loss or stealing of data and damage of data.
Data is important resource for e-commerce hence its security needs great attention to strengthen protection of e-
commerce transaction and related data [1].
4.4 Application Defence.

The application defence is an integral part of the security model to provide a strong base for the application.
When developing an e-commerce application it need to be a secure system, so the attacker can not directly attack
to the application, it defence from the attackers [1].
4.5 Establishment of Internal Control System.

The internal control system mainly consist of the organizational control system, the control system of
development and maintenance, data access control and backup system, application control system, and daily
management system. At the same time, the security audit is applied to the complete assessment of the security of
e-commerce system and the prediction of e-commerce risk in order that e-commerce are directed to improve the
safety measures, and the emergency mechanism to deal with the risk[7].
115
4.6 Personnel Training.

A personal trainer is a professional with the knowledge and experience to offer advice and guidance in the
areas of expertise. These professionals can assist user by designing a safe and effective program to help user to
achieve the organizational goals. A trainer can assist a user with different techniques to improve performance
and skills [8].
5. Introduction of Fuzzy Iteration Model

FIM (fuzzy iteration model) is evaluation method that applies composition of fuzzy relation theory and get
some quantitative factors used for quantitative analysis which includes single level and multilevel; for such a
complicated system of E-commerce multilevel fuzzy iteration model is proposed in this paper.
The step of multilevel FIM as following:
(1) Universe of discourse A is divided into Z subsets according to some attribute, recording as following:
where the Ai = {ai1, ai2, ai3, ….., aif } i = 1,2, …., z; f is factor number in Ai.
(2) Going single level FIM for every Ai assuming evaluation set B = { b1, b2, b3, …., bm }, set of factors
weighing of Ai as follows:
Xi = (x1, x2, x3, …., xif)
Evaluation matrices of single factor of Ai is Ri; evaluation model of single factor is

Xi • Ri = Bi (i = 1, 2, 3, …..f)
(3) Regarding every Ai as a factor, Bi is regarded a single factor to evaluate: in this way a membership
matrices can be obtained as following:
Ai as one part of A reflects its one attribute. According to their importance weighing set can be given as
following : X = (x1, x2, x3, ….xf) : and then second level FIM can be obtained as following:
X•R = B. Continually, three levels or multilevel model can be obtained.
(A) Establish of evaluation index system for E-commerce
116
E-commerce system includes network and information system; it is a complicated system including hardware
and software, outside influence factor and internal factor; lots of factor are inter inhibitive. Comprehensively
considering various factors, evaluation index system has been established show in figure 1.
Connection
Break A12
Phy
sical
A1 Disk break A11
Internal thread A21

Viru
s External thread A22
surv
eilla
nce Worms A23
A2
Zombie A24
Evalu
ation Data Key break A31
index defe
syste nce
m for A3 Information
E- Thrifts A32
comm
erce
A
App Password break A41
licat
ion
Roll changed A42
defe
nse
A4 Invalid
transaction A43
Pers
onal Trend user A51
train
ing Un trend user A52
A5
Figure1: Evaluation index system of E-commerce
(B) Fuzzy Iteration Model for E- commerce Security
Above FIM model is applied into E-commerce as following:
Confirming factor set A
A = {A1, A2, A3, A4, A5 } = { Physical management, virus surveillance, Data defense, Application Defense,
Personal Training } ; where, A1 = { A11, A12} = { Disk Break, connection Break}; A2 = {A21, A22,
A23, A24} = {Internal Thread, External thread, Worms, Zombie}; A3 = { A31, A32 } = { Key break,
Information Thrifts}; A4 = {A41, A42, A43} = { Password break, Roll changed, Invalid transaction}; A5 =
{A51, A52} = { Trend user, Un trend user};
Evaluation result set C = {c1, c2, c3, c4, c5} = {Very large risk, large risk, middling risk, smaller risk, very
smaller risk}
(1) First level FIM for every Ai :
Single factor evaluation matrices for Ai can be obtained by specialist appraising; weighing set Ai can be
obtained by specialist scoring; First evaluation result Bi (i= 1, 2, 3, 4) can be obtained.
(2) Second level FIM for E-commerce by specialists, weighing set A can be obtained.
117
B is second level valuation result. More level evaluation result can be continually calculated as such.
6. Conclusion
In this paper we have proposed new security risk model for risk analysis. This model based on the concept of
fuzzy logic. Using this model we find the risk factor in the field of E-Commerce and internet security
constraints. In this model
Mathematical functions are used so computational time of process is slow. In future minimize the time
complexity of FIM model.
Acknowledgement
researchers.
References
[1] Li Bo, Xu Congwei, “E-commerce Security Risk Analysis and Management Strategies of Commercial
Bank”, international forum on information technology and applications.2009, IEEE computer society.
[2] Yao Youli, Liu Jie, Jia Quan,“Risk Assessment Model for E-commerce Security based on FCE” proceeding
of the 2009 international symposium on web information system and applications (WISA’09).
[3] Limin, Fei Yaoping “A layered network evaluation method” [j]. Microcomputer developing. 2003,13(6):
106-108(chinese)
[4] Hua Liming, Wang Hua. “Multilevel fuzzy evaluation method for network security” [j]. Liaoning
Engineering Technology University Journal.2004,23(4):510-513.(chinese)
[5] Zhao Yiyao, Wang Shiping. “Security evalution model for E-commerce based on PRA” [j]. Computer
Engineering and Design.2008 29(17): 4420-4422.(chinese)
[6] Li Bo, Xu Congwei, “E-commerce Security Risk Analysis and Management Strategies of Commercial
Bank”, international forum on information technology and applications.2009, IEEE computer society.
[7] Yuanqiao Wen, Chunhui Zhou, Juan Ma, Kezhong Liu, “Research on E-Commerce Security Issues”,
international seminar on business and information management,2008, ieee computer society.
[8] Lu Tao, Lei Xue, “ Study on Security Framework in E-Commerce”, 2007 ieee
[9] Wang Liping, “Study of the Electronic Business Security Risk Management in E-Commerces”, journal of
Zhongnam university of electronic and law, 2007,(1) pp,75-78.
[10] Luis Navarrow, “Information Security Risk and Managed Security Service”, information security technical
report, 2001, 6(3) pp. 28-36.
[11] B. Fernandez-Muniz, J. M. Montes-peon, C. J. Vazquez-ordas, “Safety Management System: Development
and Validation of Multidimensional Scale”, Journal of Loss prevention in the process industries, no.20,
2007, pp. 52-68.
[12] Nie Jin, Lei Xue. “Chiness Online Banking Security Analysis”. The fifth wuhan international conference on
e-business. Volume I, 2006, pp. 662-665.
[13] China Internet Network Information Center, The eighteenth statistics reporter of the development Chinese
internet, jan. 2006.
[14] China internet network information center, The ninetheenth statistics reporter of the development Chinese
internet, jan. 2007/
118
Analysis of Bridgeless PFC Boost Converter under

CCM and DCM/CCM Boundary
M.Gopinath1, S.Ramareddy2
Research Scholar, Bharath University, Chennai, India.1
Professor, Jerusalem College Of Engg, Chennai, India.2
mgopinath_10@yahoo.co.in
Abstract
In this paper, a systematic review of bridgeless PFC boost rectifiers, also called dual boost PFC rectifiers, is
presented. Performance comparison of the bridgeless PFC boost rectifier under CCM and DCM/CCM is
performed. High-power efficiency is an important concern in the design of a switching mode power supply,
especially for energy saving and environmental protection. Mean while bridgeless PFC attracts more interest for
its ability to reduce conduction loss without the input rectifier bridge. Bridgeless PFC has been viewed as a
potentially useful technique to improve low line input efficiency; it is not certain that is where the demand for
improvement lies today. Simulation results in both CCM and DCM/CCM boundary operations are provided.
Keywords: Bridgeless power factor correction (PFC), common-mode noise, continuous-conduction mode
(CCM), discontinuous- conduction mode DCM/CCM boundary, dual-boost PFC rectifier, single-phase rectifier.
119
I. INTRODUCTION
Generally line current harmonic standards have led to a significant effort in developing AC/DC
converter. High-frequency power-factor correction (PFC) circuits, which draw a current nearly proportional to
the input voltage from the grid, have been an active research topic in power electronics. Although their typical
performance is excellent, for some large-volume applications such as household appliances and large-sized flat
panel displays, they imply an unacceptable increase in cost and complexity in the power conversion unit. These
applications require inexpensive and reliable solutions. Recently, in an effort to improve the efficiency of the
front-end PFC rectifiers, many power supply manufacturers and some semiconductor companies have started
looking into bridgeless PFC circuit topologies. Generally, the bridgeless PFC topologies, also referred to as dual
boost PFC rectifiers, may reduce the conduction loss by reducing the number of semiconductor components in
the line current path. Generally, the continuous-conduction-mode (CCM) boost converter has been widely used
for a PFC circuit. However, the operation of the CCM boost converter causes high reverse-recovery losses of the
output diode and electromagnetic interference (EMI) noises. Loss analysis for both continuous-conduction mode
(CCM) and discontinuous-conduction mode (DCM)/CCM boundary operations are provided.
Fig.01: Bridgeless PFC Boost Converter

II. BRIDGELESS PFC CIRCUIT
The bridgeless PFC circuit is shown in Fig.01. The boost inductor is split and located at the AC side
to construct the boost structure. In this half line cycle, MOSFET M1 and boost diode D1, together with the boost
inductor construct a boost DC/DC converter. Meanwhile, MOSFET M2 is operating as a simple diode. The input
current is controlled by the boost converter and following the input voltage. During the other half line cycle,
circuit operation as the same way. Thus, in each half line cycle, one of the MOSFET operates as active switch
and the other one operates as a diode: both the MOSFETs can be driven by the same signal.
Fig.02: Circuit diagram of CCM bridgeless PFC boost rectifier.
Comparing the conduction path at every moment, the bridgeless PFC inductor current only goes through
two semiconductor devices, but inductor current goes through three semiconductor devices for the conventional
PFC circuit. The bridgeless PFC uses one MOSFET body diode to replace the two slow diodes of the
conventional PFC. Since both the circuits operating as a boost DC/DC converter, the switching loss should be
the same. Thus the efficiency improvement relies on the conduction loss difference between the two slow diodes
and the body diode of the MOSFET. Besides, comparing with the conventional PFC, the bridgeless PFC not only
reduces conduction loss, but also reduces the total components count.
120
Fig.03: Circuit diagram of DCM/CCM boundary bridgeless PFC boost rectifier.
Key components of the CCM and DCM/CCM boundary bridgeless PFC boost rectifiers are used in Fig.02 and
Fig.03, respectively. As can be seen in Fig.02 and Fig.03, diodes and are implemented with the two bottom
diodes of the full-bridge rectifier D15XB60. In the corresponding conventional PFC boost rectifiers, the whole
full-bridge rectifier D15XB60 is used for diodes whereas, for boost inductor only one boost inductor in Fig.02
and Fig.03 is employed. The input filter and the bulk capacitor in the conventional PFC boost rectifiers are
identical to those in the bridgeless PFC boost rectifiers. For the boost switches, the IPP60R099CS MOSFETs
were used. The major loss components in both CCM and DCM/CCM boundary implementations are the input
diodes and the boost switches. In CCM implementation, the loss of the EMI filter is also significant. The
efficiency improvement is 0.46% in CCM implementation and 0.69% in DCM/CCM boundary implementation.
Generally, the efficiency could be further improved by using MOSFETs with lower on-resistance. If the
IPP60R099CS or the IPW60R044CS MOSFET is used for the boost switches, the efficiency of the bridgeless
PFC boost rectifier is higher than that of the conventional PFC boost rectifier. The employment of the
IPP60R099CS MOSFET yields the most uniform high efficiency in the whole output power range. It can be
concluded that the IPP60R099CS MOSFET used in the bridgeless PFC boost rectifier operating at the
DCM/CCM boundary, exhibits an optimal balance between the conduction losses and the switching losses in the
output power range. Both modes of operation have advantages and disadvantages. The main disadvantage in
using CCM is the inherent stability problems. However, the switch and output diode peak currents are larger
when the converter is operating in the DCM mode. Larger peak currents necessitate using larger current and
power dissipation rated switches and diodes. Also, the larger peak currents cause greater EMI/RFI problems.
Most modern designs use CCM because higher power densities are possible.
III. SIMULATION RESULTS:

Simulation is done using Matlab Simulink and the results are presented. Bridgeless boost
converter is shown in Fig.04.a. It is assumed that a controlled switch is implemented as the power MOSFET
with its inherently slow body diode. Ac input voltage and current are shown in Fig.04.b.It can be seen that the
current and voltage are almost in phase. DC output current and output voltage are shown in Fig.04.c and Fig.04.d
respectively.
Fig 4.a: Bridgeless Boost converter
121
Fig 4.b: AC Input Voltage and Current Fig 4.c: DC Output Current
Fig 4.d: DC output voltage
The simulation result clearly implies that the power factor correction is done at input side, the number of
components is reduced during conduction. Therefore the boosted output voltage has better efficiency.
A constant switching frequency CCM simulation circuit and a variable switching frequency
(DCM/CCM boundary simulation circuit were built for the universal ac-line input. The circuit diagram of the
CCM and DCM/CCM boundary bridgeless PFC experimental circuit is shown in Fig.05.a and Fig.06.a
respectively. Because of cost concerns and high light-load efficiency requirements, in both experimental circuits,
the MOSFET was used as the boost switch. Ac input voltage and current of the CCM are shown in Fig.05.b. DC
output voltage of CCM is shown in Fig.05.c respectively.
122
Fig.05.a: CCM Simulation Circuit Fig.05.b: Ac input voltage and current under CCM
Fig.05.c: DC output voltage under CCM Fig.06.a: DCM/CCM boundary Simulation Circuit
The measured AC-line input voltage and current of DCM/CCM boundary are shown in Fig.06.b. DC output
voltage of DCM/CCM boundary is shown in Fig.06.c respectively. The current wave form is shown in
Fig.06.d.It was found that line current value is better in CCM when compare to DCM/CCM boundary
implementation.
Fig.06.b: Ac input voltage and current under

DCM/CCM boundary
123
Fig.06.c: DC output voltage under DCM/CCM
IV -CONCLUSION
Bridgeless PFC Converter is modeled and simulated using Matlab. CCM and DCM/CCM boundary circuits
are developed and they are used successfully for simulation. The simulation studies indicate that the power factor
is nearly unity by employing the Bridgeless boost converter. This converter has advantages like reduced
hardware, high performance and improved power factor. The simulation results are in line with the predictions. It
was found that the DCM/CCM boundary implementation had a slightly better efficiency by 0.8% at full load and
by almost 5% at 20% load at worst case than the CCM implementation. However, the quality of the line current
in the CCM implementation was slightly better than that in the DCM/CCM boundary implementation. The
hardware implementation will be done in future
Fig.06.c: Current waveform under DCM/CC

REFERENCES
[1]. W.-Y. Choi, J.-M. Kwon, E.-H. Kim, J.-J. Lee and B.-H. Kwon, “Bridgeless boost rectifier with low
conduction losses and reduced diode reverse-recovery problems,” IEEE Trans. Ind. Electron., vol. 54, no. 2,
pp. 769–780, Apr. 2007.
[2]. H. Ye, Z. Yang, J. Dai, C. Yan, X. Xin, and J. Ying, “Common mode noise modeling and analysis of dual
boost PFC circuit,” in Proc. Int.Telecommunication Energy Conf., Sep. 2004, pp. 575–582.
[3]. B. Lu, R. Brown, and M. Soldano, Bridgeless PFC implementation using one cycle control technique,” in
Proc. IEEE Applied Power Electronics Conf., Mar. 2005, pp. 812–817.
[4]. P. Kong, S.Wang, and F. C. Lee, “Common mode EMI noise suppression in bridgeless boost PFC
converter,” in Proc. CPES Power electronics Conf., Apr. 2006, pp. 65–70.
[5]. Huber L., Yungtaek Jang, Jovanovic M.M.: Performance Evaluation of Bridgeless PFC Boost Rectifiers.
IEEE Transactions on Power Electronics, 2008, Vol 23 no3, pp. 1381–1390.
124
[6] Chu, G.; Tse, C.K.; Siu Chung Wong, “Line-Frequency Instability Of PFC Power Supplies”, IEEE Trans.,
vol. 24, pp. 469-482, 2009.
[7] Weiming Lin etal, "A high efficiency Zero-Voltage-Switching Boost Voltage doubler With Low Conduction
Losses," IEEE APEC2008.
[8] Prodic A, Maksimovic D, Erickson R W.: Dead-zone digital controllers for improved dynamic response of
low harmonic rectifiers, IEEE Transactions on Power Electronics, 2006, Vol 21 no1, pp. 173-181.
[9] Aleksandar P.: Compensator design and stability assessment for fast voltage loops of power factor correction
rectifiers, IEEE Transactions on Power Electronics, 2007, Vol 22 no5, pp. 1719-1730
[10] Y. Jang and M M. Jovanovic, A Bridgeless PFC Boost Rectifier with Optimized Magnetic Utilization, in
IEEE Trans. on Power Electronics, Vol. 24, No.1. Jan. 2009, pp. 85-93.
[11] Texas Instruments, UCC28060EVM 300-W Interleaved PFC Pre Regulator, SLUU280B-May 2007-
Revised July 2008.
[12] S.Wang, F. C. Lee, and J. D. vanWyk, “Design of inductor winding capacitance cancellation for EMI
suppression,” IEEE Trans. Power Electron., vol. 21, no. 6, pp. 1825–1832, Nov. 2006.
About Authors
M. Gopinath has obtained his B.E degree from Bharathiar University, Coimbatore in the year
2002. He obtained his M-Tech degree from Vellore Institute of Technology, Vellore in the year
2004.He is presently doing his research work at Bharath University, Chennai. He is working as
an Assistant Professor/EEE, at Ganadipathy Tulsi’s Engg College, Vellore.
S. Ramareddy is Professor and Head of Electrical Department Jerusalem Engineering

College, Chennai. He obtained his D.E.E from S.M.V.M Polytechnic, Tanuku, A.P. A.M.I.E in
Electrical Engg from institution of Engineers (India), M.E in Power System Anna University.
He received Ph.D degree in the area of Resonant Converters from College of Engineering,
Anna University, Chennai. He has published over 20 Technical papers in National and
International Conference proceeding/Journals. He has secured A.M.I.E Institution Gold medal
for obtaining higher marks. He has secured AIMO best project awards and Vijaya Ratna
Award. He has worked in Tata Consulting Engineers, Bangalore and Anna University, Chennai. His research
interest is in the area of resonant converter, VLSI and Solid State drives. He is a life member of Institution of
Engineers (India), Indian Society for India and Society of Power Engineers. He is a fellow of Institution of
Electronics and telecommunication Engineers (India). He has published books on Power Electronics and Solid
State circuits.
125
IJCIIS Reviewers
A. Govardhan, Jawaharlal Nehru Technological University, India
Ajay Goel, Haryana Institute of Engineering and Technology, India
Akshi Kumar, Delhi Technological University, India
Alok Singh Chauhan, Ewing Christian Institute of Management and Technology, India
Amandeep Dhir, Helsinki University of Technology Finland, Denmark Technical University, Denmark
Amol Potgantwar, Sandip Institute of Technology and Research Centre, India
Aos Alaa Zaidan Ansaef, Multimedia University, Malaysia
Arul Lawrence Selvakumar, Kuppam Engineering College, India
Ayyappan Kalyanasundaram, Rajiv Gandhi College of Engineering and Technology, India
Azadeh Zamanifar, Iran University of Science and Technology University and Niroo Research Institute, Iran
Bilal Bahaa Zaidan, University of Malaya, Malaysia
B. L. Malleswari, GNITS, India
B. Nagraj, Tamilnadu News Prints and Papers, India
C. Suresh Gnana Dhas, Vel Tech Multitech Dr.Rengarajan Dr.Sagunthla Engg. College, India
C. Sureshkumar, J. K. K. M. College of Technology, India
Deepankar Sharma, D. J. College of Engineering and Technology, India
Durgesh Kumar Mishra, Acropolis Institute of Technology and Research, India
D. S. R. Murthy, SreeNidhi Institute of Science and Technology, India
Hafeez Ullah Amin, KUST Kohat, NWFP, Pakistan
Hanumanthappa Jayappa, University of Mysore, India
Himanshu Aggarwal, Punjabi University, India
Jagdish Lal Raheja, Central Electronics Engineering Research Institute, India
Jatinder Singh, UIET Lalru, India
Iman Grida Ben Yahia, Telecom SudParis, France
Leszek Sliwko, CITCO Fund Services, Ireland
M. Azath, Anna University, India
Md. Mobarak Hossain, Asian University of Bangladesh, Bangladesh
Mohammed Salem Binwahlan, Hadhramout University of Science and Technology, Yemen
Mohamed Elshaikh, Universiti Malaysia Perlis, Malaysia
M. Thiyagarajan, Sastra University, India
Nahib Zaki Rashed, Menoufia Univesity, Egypt
Nagaraju Aitha, Vaagdevi College of Engineering, India
Natarajan Meghanathan, Jackson State University, USA
N. Jaisankar, VIT University, India
Ojesanmi Olusegun Ayodeji, Ajayi Crowther University, Nigeria
Oluwaseyitanfunmi Osunade, University of Ibadan, Nigeria
Perumal Dananjayan, Pondicherry Engineering College, India
Piyush Kumar Shukla, University Institute of Technology, Bhopal, India
Poonam Garg, Institute of Management Technology, India
Praveen Ranjan Srivastava, BITS, India
Rajesh Kumar, National University of Singapore, Singapore
Rakesh Chandra Gangwar, Beant College of Engineering and Technology, India
Raman Kumar, D A V Institute of Engineering and Technology, India
Raman Maini, University College of Engineering, Punjabi University, India
Sateesh Kumar Peddoju, Vaagdevi College of Engineering, India
Shahram Jamali, University of Mohaghegh Ardabili, Iran
Sriman Narayana Iyengar, India
Suhas Manangi, Microsoft, India
Sujisunadaram Sundaram, Anna University, India
Sukumar Senthilkumar, National Institute of Technology, India
S. S. Mehta, J. N. V. University, India
S. Smys, Karunya University, India
S. V. Rajashekararadhya, Adichunchanagiri Institute of Technology, India
Thipendra P Singh, Sharda University, India
T. Ramanujam, Krishna Engineering College, Ghaziabad, India
Vishal Bharti, Dronacharya College of Engineering, India
V. Umakanta Sastry, Sreenidhi Institute of Science and Technology, India
126

Ijciis July 2010 Vol. 1 No. 5

Diunggah oleh

Informasi Dokumen

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Ijciis July 2010 Vol. 1 No. 5

Diunggah oleh

Hak Cipta:

Format Tersedia

International Journal of

Computational Intelligence and

IJCIIS Editor and Publisher

Publishing Date: July 30, 2010

Members of IJCIIS Editorial Board

Prof. A Govardhan, Jawaharlal Nehru Technological University, India

Journal Website: https://sites.google.com/site/ijciisresearch/

1. A Secure Architecture Platform for Mobile Electronic Payment (pages 4-9)

A Secure Architecture Platform for Mobile Electronic Payment

M. Udayapal Reddy1, R.K. Pateriya2, R.K. Baghel3

Keywords: SMEP, symmetric key, RSA, AES, wireless mesh network.

4. Security requirements of Wireless Mesh Networks (WMN)

5. Secure Mobile Electronic Payment (SMEP) model Platform

5.1. The key Distribution Procedure

5.2. Payment Transaction Procedure

Towards Improving E-Registration System Using Email Technology

Hamdan.O.Alanazi 1, Rami Alnaqeib 1, M.A.Zaidan 1, A.A.Zaidan 2, B.B.Zaidan 2,

Keyword: - E-Registration, Knowledge Management, Elecronic Email Technology.

3. E-Registration and Knowledge Management Parts

5. Conjunctive Relations and the Synergic Solution

Figure 1: KM Framework of FSKTM E-Registration Online Solution

ii. System E-mail Structure

iii. Concept of E-mail

Figure 2: Concept of E-mail connection

iv. System E-mail Programs

v. System E-mail Addresses

EFSM-Editor: An editor for fuzzy semantic data models

Noura Labidi, Lassaad Ellouze and Rafik Bouaziz

MIRACL, Multimedia InfoRmation systems and Advanced Computing Laboratory

noura.labidi@yahoo.fr; las_ellouze@yahoo.fr; raf.bouaziz@fsegs.rnu.tn

The d.o.m of a fuzzy entity e in a fuzzy class K is then calculated as follows:

With  p i ( i ) is the partial membership function relating to the property pi.

4. Development of a graphical editor for EFSM according to MDA approach

Figure 1: Extract of the diagram relative to MS disease under EFSM [5]

Management of an EFSM Management of an EFSM

Figure 2: Use case diagram modeling the requirements of EFSM-Editor

1.2. The computational models of EFSM-Editor

Figure 3: The CIM of EFSM-Editor

Figure 4: First PIM of EFSM-Editor

(or number of links between each two types of figurines).

Figure 5: Second PIM of EFSM-Editor

Figure 6: Screenshot of EFSM-Editor

Figure 7: Extract of XMI code generated relative to SEP model [5]

Credit Card Fraud Detection using Kernel in SVM

3. SVM (Support Vector Machine)

3.1 SVM classification

Alternatively, it could also be implemented using

Step 6: Train SVM.

4.2 Proposed model for fraud detection

Input transaction probability Select SVM kernel

Generated synthetic data

Input false transaction

Categorized data Generated classification

Current transaction data Decision equation

Figure 1: Fraud detection model

7. Conclusion and Future Scope

Attribute vector optimization for a non supervised classification by

Random generation of an initial

Random generation of an initial population

Figure 2: Standard SE algorithm.

Fix the number of classes C

6. Classes number optimization

TV  1 Rit V(m) Rit V(m) 

This criterion is to be maximized.

Arutchelvan.G, Dr. Srivatsa.S.K and Dr. Jagannathan.R