Anda di halaman 1dari 7


discussions, stats, and author profiles for this publication at:

Detecting evolutionary financial statement


Article in Decision Support Systems · February 2011

DOI: 10.1016/j.dss.2010.08.007 · Source: DBLP


40 878

2 authors, including:

Wei Zhou
ESCP Europe


Some of the authors of this publication are also working on these related projects:

Big Data & Business Analytics View project

All content following this page was uploaded by Wei Zhou on 24 September 2015.

The user has requested enhancement of the downloaded file.

Decision Support Systems 50 (2011) 570–575

Contents lists available at ScienceDirect

Decision Support Systems

j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / d s s

Detecting evolutionary financial statement fraud

Wei Zhou a,⁎, Gaurav Kapoor b
Information Systems and Technologies, ESCP Europe, 75543 Paris cedex 11, France
Information Systems and Operations Management, University of Florida, Gainesville, Florida 32611, USA

a r t i c l e i n f o a b s t r a c t

Available online 24 August 2010 A fraudulent financial statement involves the intentional furnishing and/or publishing of false information in it
and this has become a severe economic and social problem. We consider Data Mining (DM) based financial
Keywords: fraud detection techniques (such as regression, decision tree, neural networks and Bayesian networks) that
Financial statement fraud help identify fraud. The effectiveness of these DM methods (and their limitations) is examined, especially when
Data mining technique new schemes of financial statement fraud adapt to the detection techniques. We then explore a self-adaptive
Neural networks framework (based on a response surface model) with domain knowledge to detect financial statement fraud.
We conclude by suggesting that, in an era with evolutionary financial frauds, computer assisted automated
fraud detection mechanisms will be more effective and efficient with specialized domain knowledge.
© 2010 Elsevier B.V. All rights reserved.

1. Introduction misapplication of accounting principles, policies, and procedures used

to measure, recognize, report, and disclose business transactions [20].
Since the booming of the Internet and the invention of other Many techniques based on data mining have been investigated
modern technologies, there has been a dramatic increase in fraudulent and implemented to detect financial statement fraud, including
schemes associated with all facets in the business world. Some of these regression, decision trees, neural networks and Bayesian belief
commonly observed schemes include credit card fraud, financial networks [12]. These techniques have been shown to be successful
statement fraud, e-commerce transaction fraud, insurance fraud, in their early stages. However, there is no agreement on which data
money laundering, computer intrusion fraud, telecommunications features and techniques are best for detection. Also, while supervised
fraud, and subscription fraud. Statistic and machine learning based learning techniques have been among the dominant methods used
technologies have been shown to be an effective way to deter and for detecting financial statement fraud, a majority of related
detect fraud, but fraudsters are adaptive and are usually able to find implementations do not keep track of new variations in the methods
ways to circumvent them. Existing fraud detection techniques for most designed for committing fraud. Moreover, financial fraud is becoming
of the situations involving fraud usually share very similar data mining more and more difficult to detect using the current detection
principles, but they can differ in many aspects with specialized domain techniques. A CEO who is truly knowledgeable and wants to really
knowledge [5]. commit a crime has the necessary resources to easily outwit the
Financial statement fraud in particular has cast rapidly increasing system and is able to fool any detection mechanism [7].
adverse impact not only on individual investors but the overall Despite the increased of time and effort that has been spent to
stability of global economies. Although there are minor variations in detect the same, the number of detected frauds1 and the detection
its definition, a financial statement fraud is defined by the Association rate2 have largely decreased [7]. When the executives who are
of Certified Fraud Examiners as “The intentional, deliberate, misstate- involved in financial fraud are well aware of the fraud detection
ment or omission of material facts, or accounting data which is techniques and software, which are usually public information and
misleading and, when considered with all the information made are easy to obtain, they are likely to adapt the methods in which they
available, would cause the reader to change or alter his or her commit fraud and make it difficult to detect the same, especially by
judgment or decision.” In practice, financial statement fraud might existing techniques. There exists an urgent need for new methods that
involve: (1) manipulation of financial records, (2) intentional is not only efficient but effective to catch up with these probable
omission of events, transactions, accounts, or other significant newly emerged or adaptive financial shenanigans. We (1) consider
information from which financial statements are prepared, or (3) existing detection techniques based on data mining, (2) provide an

In 2003, the SEC issued 77 financial statement fraud AAERs (Accounting and
Auditing Enforcement Releases) relating to its registrants. It was 44 in 2006.
⁎ Corresponding author. Average time elapsed between the initiation date and the date the SEC issued
E-mail addresses: (W. Zhou), (G. Kapoor). AAERs increased by one-third, to 5.6 years in 2006 from 4.1 years in 2001.

0167-9236/$ – see front matter © 2010 Elsevier B.V. All rights reserved.
W. Zhou, G. Kapoor / Decision Support Systems 50 (2011) 570–575 571

overview of existing financial shenanigans and their trend, and (3) The objective of decision trees is classification by dividing
suggest a new framework to detect evolutionary financial statement observations into mutually exclusive and exhaustive subgroups by
fraud. properly selecting attributes that best separate the sample. Koh and
The remainder of this paper is organized as follows. We review the Low [13] construct a decision tree to predict the hidden problems in
application of regression, decision trees, neural networks and financial statements by examining the following six variables: quick
Bayesian belief networks in financial statement fraud detection in assets to current liabilities, market value of equity to total assets, total
the next section. We survey the history and trend of contemporary liabilities to total assets, interest payments to earnings before interest
financial fraud in Section 3. We analyze the effectiveness and and tax, net income to total assets, and retained earnings to total assets.
limitation of existing fraud detection technologies in Sections 4 and The above mentioned data mining techniques have generally been
5 and then suggest a framework that addresses the problem when shown to be effective in detecting financial statement fraud. However,
emerging financial fraud is evolutionary. Section 6 concludes the they are not without limitations. For example, while these techniques
paper with a brief discussion on the insights garnered and possible are well developed for predictive modeling, they are not as well
future research. developed for effect assessment. In particular, test statistics for
assessing the effects of independent variables on dependent variables
2. Review on detection techniques have not yet been constructed for some data mining techniques.
Incidentally, this shortcoming also demands with the challenge to
Classification has been the most popular and the only way used so develop more effective mechanisms especially in an adaptive
far to identify fraudulent financial statements [8]. Most financial economic environment where financial fraudsters learn to circumvent
statement fraud (FSF) auto-detection programs use supervised existing automated detection systems.
machine learning methodologies [1,3,4,9–11,14,18,21,22] that usually
have a two-stage procedure, where in the first stage a model is trained 3. Detection with domain knowledge on FSF
by using a training sample. The training sample is organized in tuples
and attributes, with the class label attribute containing values Financial statement fraud, including motivations, opportunities,
indicating the pre-defined class to which each tuple belongs. In the and rationalizations for management to commit such fraud, has been
second stage, objects are classified through the model obtained from extensively studied by researchers in finance. Loebbecke et al. [15]
the first stage. After reviewing relevant research in data mining-based suggest a model consisting of three variables that may explain
financial statement fraud detection literature, we observe that the financial statement fraud: (C) the degree to which conditions are such
following five methods have been used so far in this general area. that a financial fraud could be committed, (M) the degree to which the
These methods include regression, decision trees, neural networks, management has a reason or motivation to commit financial fraud,
Bayesian networks and support vector machines. and (A) the degree to which the management has an attitude or set of
Regression is the most widely used method to detect financial ethical values such that they would allow themselves to commit
statement fraud [1,3,4,11,18,21,22]. Transformations of variables in management fraud. These three variables together form the assess-
regression models have also been studied in the context of fraud ment model such that the possibility of having financial statement
detection, including logit, stepwise-logistic, multi-criteria decision aid fraud (FSF) can be described as a function.
method and exponential generalized beta two. For example, Spathis
[21] used a collection of data from 76 firms that include 38 fraudsters P ðFSF Þ = f ðC; M; AÞ ð1Þ
and 38 non-fraudulent firms in Greece. They use ten financial
variables and logistic multivariate regression to identify the relation- where if C or M or A = 0, then P(FSF) = 0.
ship among factors associated with financial statement fraud. A total According to Rezaee [19], fraud accomplishment can be explained
of ten financial ratios such as the net profit to total assets ratio, the by three variables: (1) conditions, (2) corporate structure and (3)
ratio of total debt to total assets, financial distress, the inventories to choice. The sufficient incentives and opportunities for a company to
sales ratio, and the working capital to total assets ratio are selected for commit financial statement fraud can be interpreted by the pattern
examination as potential predictors of FSF. The results indicate that exhibited by these three variables. We, however, believe that certain
companies with high inventories with respect to sales, high debt to combination of the variables mentioned in this model also leads to
total assets, low net profit to total assets, low working capital to total certain suitable fraud strategies and, by measuring the pattern among
assets and high financial stress are more likely to manipulate financial the variables and integrating the findings in auto-detection heuristics,
statements. we should be able to pinpoint the unique shenanigans and their
A neural network is another popular data mining technique that underlying dynamic to commit fraud.
has been successfully used to detect financial statement fraud We first review possible variables that can be utilized in an auto-
[6,9,14,16,24]. Neural network doesn't assume an attribute's inde- detection system. In Rezaee's[19] 3C's model as shown in Fig. 1,
pendence and is capable of mining inter-correlated data and is a “conditions” refers to the economic and financial pressures that a
suitable alternative for problems where some of the assumptions corporation faces. Financial pressures, such as pressure to meet
associated with regression are not valid. White [24], nonetheless, has analysts' earning estimates, can be a key factor stimulating earnings
shown that feed-forward neural networks, which require no pre- management and resulting in financial statement fraud. The principle
specified functional form, perform the same stochastic approximation underlying this variable is that financial statement fraud will most
as nonlinear regression. Back propagation neural network allows the probably occur if the benefits for fraudulent management outweigh
network to adapt and has become one of the most popular techniques the associated costs. Management compares the benefit, in terms of
for prediction and classification problems. The back propagation possible increase in the company's stock price or the possible savings
learning process works in small iterative steps that continuously make related to preventing share price from decreasing, with the possible
small changes to the weights in each neural network layer, which are cost of committing fraud in terms of probability and consequences of
calculated to reduce the systematic error. The iteration is repeated detection. Financial pressures, such as the inability to meet analysts'
until the overall error value drops below some pre-determined earning estimates or decline in quality and quantity of earnings, are
threshold [13]. Drawbacks of implementing neural networks to often motivations for management commitment in financial frauds.
discover FSF is that neural network is not accurate if the data is “Capital structure” refers to the existence of an effective corporate
volatile or if the causal functionality evolves in a direction that is not governance mechanism (such as internal control structure and audit
pre-defined. committees) that could discourage management from committing
572 W. Zhou, G. Kapoor / Decision Support Systems 50 (2011) 570–575

released from January 2000 to December 2006 and sorts each company
into one of the following nine industries: aviation and transport
services manufacturing; consumer business; public sector; energy and
resources; real estate; financial services; technology, media, and
telecommunications; life sciences; and health care. Doloitte [7] reports
that the number of financial statement frauds that the SEC issued has
decreased from 77 in 2003 to 44 in 2006. Although the number of
reported frauds has decreased dramatically after the Sarbanes–Oxley
Act was introduced in 2002, it is arguable whether the real number has
decreased by that much. In reality, we have seen more fraudulent
behaviors since 2008 when the economy has been going towards the
lowest point since the great depression in 1929.
It leads us to believe that fraud types and industry patterns
changed over time. It is important to understand how fraud schemes
have evolved and it is more important to predict by any means
possible, the direction of the change and keep the automatic fraud
detection techniques up-to-date. The findings may have significance
Fig. 1. Rezaee's 3C financial statement fraud model. to develop more robust business processes as well as adaptive fraud
detection mechanism for managing/deterring/detecting the risk of
fraud. The role of corporate governance devices can also be discussed fraud. The efficacy of such processes will depend on knowing the fraud
in relation to other social and economic characteristics of different schemes typically committed and industry-by-industry differences in
countries where fraud can be committed. It is also important to those schemes.
explain to what extent an effective corporate governance system can A majority of data mining-based financial statement fraud
help to prevent and detect fraud. detection techniques are primarily based on classification, which is
“Choices” refer to the management's option, which could either be the process of explaining and differentiating data classes in order to
ethical strategies of continuous improvements of earnings or illegal predict the class of objects whose class label is unknown. As compared
earning manipulations, to deal with various financial situations. to association rule mining that deals with existing data items, the
Financial statement fraud is simply one of the choices. Regardless of classification rule mining deals with attributes and its values, clusters
the corporate structure or environmental pressure, management the attributes, and uses time-series mining or outlier detection to
could use financial statement fraud simply as a strategic tool recognize the new mode for financial statement fraud detection.
according to its own characteristics in terms of aggressiveness or Despite the early success of these pioneering FSF auto-detection
lack of moral principles. systems and algorithms, the rate of successful detection has
Both the CMA model [15] and the 3Cs model [19] mentioned above continuously decreasing over the past several years. Although it is
explain the financial statement fraud by examining selected para- certainly possible that ethical standards have greatly improved since
meters, such as the motivations, conditions of pressure, corporate 2000, given the fact that the severity and quantity of law suits for
structure, management's attitude and the choices. These variables financial statement frauds have increased recently, we are more
may each function separately, and careful selection and combination inclined to believe that managements have adapted certain ways to
of variables, however, are more likely to help explain the degree and avoid being identified by automated detection systems.
pattern of financial statement fraud, thus providing insightful and Although according to the No-Free-Lunch theorem, existing data
immediate reference for an FSF auto-detection system. mining search techniques, including random search, should perform
The strategy that the management utilizes to commit financial equally well on average, some techniques would have more accurate
statement fraud is influenced by the economic circumstances as well as estimation than others when appropriate domain knowledge can be
many other variables that have been taken into consideration by the identified and integrated. In the next section, we propose a
fraudster. Schilit [20] provides an extensive examination of the framework that utilizes domain knowledge to facilitate the detection
financial statement fraud techniques and categorizes them into of financial statement fraud in a constantly evolving economic
seven groups: recording revenue too soon or of questionable quality, environment.
recording bogus revenue, boosting income with one-time gains,
shifting current expenses to a later or earlier period, failing to record 5. Adaptive FSF detection framework and methodologies
or improperly reducing liabilities, shifting current revenue to a later
period, and shifting future expenses to the current period as a special 5.1. Framework to detect adaptive financial fraud
charge. The COSO report [2], by analyzing 204 cases of fraud presented
in the SEC's Accounting Auditing Enforcement Releases (AAERs) from Similar to traditional classification procedure, in general we consider
1987 to 1997, lists common financial statement fraud techniques in the two stages in our framework (Fig. 2). In the first stage, relevant external
following categories: improper revenue recognition, overstatement of and internal variables that differentiate the industries, economic
assets other than accounts receivable, understatement of expenses/ conditions, management's choice, timing considerations and any
liabilities, misappropriation of assets, inappropriate disclosure and other factors that have the potential to form domain knowledge are
other miscellaneous techniques. selected and experimented. In the second stage, the financial data of the
firm in question is analyzed based on this domain knowledge learned
4. Limitations of current detection techniques from the previous step. Certain detection strategy is formed accordingly
and the data is further analyzed using data mining techniques.
Despite increasingly stringent legislation such as the Foreign To realize the general model described above and to catch the real
Corrupt Practices Act and the Sarbanes–Oxley Act aimed at combating world dynamics and possible new methods to commit financial
fraud – and despite increasingly growing number of fraud auto- statement fraud, we propose an adaptive learning framework (Fig. 3).
detection systems – financial statement fraud is becoming a more and We consider, but are not limited to, exogenous parameters described
more severe public concern. Doloitte [7] reviews around 1300 in existing literature, such as capital structure, conditions, choices,
Accounting and Auditing Enforcement Releases (AAER) that the SEC management attitudes, etc. The mechanism works as follows: based
W. Zhou, G. Kapoor / Decision Support Systems 50 (2011) 570–575 573

Fig. 2. Proposed two-stage framework.

on external and internal economical circumstances, management by a list of several input variables, which may or may not be complete
chooses their action on whether or not to commit financial statement and are called independent variables. Independent variables are
fraud at the year end. Financial data and statements of concern are subject to the control of the experiment designer. Approximation of
audited and examined by a fraud detection unit. Resulting audit the relationship between the response and independent variables can
reports are further evaluated and learned by a self-adaptive module to be visualized by RSM, which consists of three factors: (1) the
collect relevant patterns and trends of each company in different experimental strategy for exploring the space of independent
industries. In the meanwhile, an adaptive fraud discovery module variables, (2) empirical statistical modeling to develop an appropriate
keeps evolving with exogenous parameters to discover unknown but approximating relationship between the response and the indepen-
possible pattern of financial statement fraud. New discoveries were dent variables, and (3) optimization methods for finding the values of
also evaluated and learned to prepare the knowledge base for future the process variables that produce desirable values of response.
fraud detection. Theoretically, the appropriate approximating model between the
To increase detection relevancy and to reduce computational response y and independent variables x1, x2,…xn can be constructed as
complexity, we also propose adaptive feature selection that fits y = f(x1, x2,…xn) + ε, where the form of the true real-time response
specific domain to choose the proper parameters for companies with function f is unknown and may be complex. ε, which usually includes
similar internal and external financial environment [13]. Once relevant measurement error on response and other unpredictable noise, is a
parameters are selected, we opt to proper methodology and data random term that represents the variability not caught by the
mining technique to detect financial fraud that evolves. As discussed in response function. If we assume that it has a normal distribution
the previous section, no single data mining-based FSF detection with zero mean, Y = E(y) = f(x1, x2,…xn). With the form of the true
technique is perfect and each of them is subject to its own handicaps. response function f kept unknown, the designer has to approximate it
We propose response surface methodology to construct the founda- and further utilize it to locate the possible response with discovered
tion in order to find the right data mining-based detection technique. independent variables.
Feature selection can be implemented to identify a relevant list of
5.2. Adaptive financial fraud detection with RSM variables that the designer may further utilize to construct the
response surface. Once we have a firm in question, we are able to find
Response Surface Methodology (RSM), a method for constructing the possibility of certain financial statement frauds based on the
global approximations to system behavior based on results calculated response surface estimation. Then, we select data mining techniques
at various points in the design space [23], is a natural fit to estimate that suit the profile that we have learned from the previous step.
the relationships between the variables and the financial statement RSM provides statistical tools for analysis of historical data and
fraud techniques. RSM provides statistically validated predictive selection of variables aimed at better prediction. The objectives for
models that can be manipulated for finding the probability of different using RSM in the context of financial statement fraud detection are to
forms of possible financial statement frauds. find the optimum response and to understand how the response
RSM is extensively applied in various situations where output changes in a given direction by adjusting the design variables [17].
performance or service quality, which is called response, is influenced When there are constraints on the design data, then the variable

Fig. 3. FSF detection strategy selection to detect adaptive financial.

574 W. Zhou, G. Kapoor / Decision Support Systems 50 (2011) 570–575

selection and experimental design has to meet requirements of the 6. Conclusion

constraints. In general, the response surface can be visualized
graphically and the graph, if in fewer than three dimensions, is helpful In recent years, data mining has gained widespread attention and
to navigate over the response surface to reach the desired outcome. increasing popularity in the financial world. Successful data mining
In a simple form, a function f (x1, x2) can be plotted versus the levels applications have been reported and recent surveys have found that
of x1 and x2, and this three-dimensional graph forms a response surface data mining has grown in usage and effectiveness. Professional
plot (Fig. 3). This figure shows an adaptive learning framework to accounting bodies have also identified data mining as an important
detect evolutionary financial statement frauds using a response technology for the new century. Implementing straightforward data
surface method (Fig. 4). We proceed by first selecting relevant mining techniques to discover financial statement fraud, however, has
variables that could either be the three parameters in the CMA many disadvantages and limits of usage. After exhaustive literature
model [15], the parameters in a 3C's model [19], a mixture from both review, we find that a majority of existing data mining techniques to
models, or any other parameters that have some causal relationship to detect financial fraud have their domain of usage and limitations.
financial statement frauds. The selected variables are denoted as Furthermore, when financial statement fraudsters have found ways
to circumvent the automatic detection programs, there is an urgent
need for a mechanism that is able to learn and use the industrial domain
Va ; Vb ; :::Vn ð2Þ
knowledge to facilitate the data mining techniques. We propose such a
framework that is based on the response surface method to
respectively, so the probability of committing financial statement automatically pivot the program in accordance with the unique
fraud in the form of k can be described as Eq. (2) that is subject to Eqs. circumstances of the firm in question. Unlike traditional financial
(3) and 4(4). We assume that the first n forms of financial statement fraud detection techniques that are based on historical financial data,
fraud have been discovered and that the number of possible forms is we further propose an innovative way, the active discovery module that
infinity, such that evolves ahead of possible fraudsters. Preparing the intelligent detection
system in anticipation before any unknown or future fraud happens
enables us to effectively detect financial statement frauds that adapt.
P ðFSFk Þ = f ðVa ; Vb ; :::Vn Þ ð3Þ Future research is needed to design the active discovery module that
is both effective and efficient. Furthermore, although we have suggested
which is subject to the response surface method to extract the domain knowledge and to
adaptively learn the changes from fraudsters, there may exist
alternatives that have equal or better performance than the RSM.
∑ P ðFSFk Þb1: ð4Þ Research is also needed to examine the circumstances under which our
k=1 suggested framework performs better than other techniques.

Fig. 4. Adaptive learning and detecting financial statement fraud using a response surface method.
W. Zhou, G. Kapoor / Decision Support Systems 50 (2011) 570–575 575

References [15] J.K. Loebbecke, M.M. Eining, J.J. Willingham, Auditors' experience with material
irregularities: frequency, Nature, and Detectability, Auditing: A Journal of Practice
[1] M.S. Beasley, An empirical analysis of the relation between the board of director and Theory 9 (1989) 1–28.
composition and financial statement fraud, The Accounting Review of Finance 71 [16] M.D. Odom, R. Sharda, A neural network model for bankruptcy prediction, IJCNN
(4) (1996) 443–465. International Joint Conference on Neural Networks, 1990, pp. 163–168.
[2] M.S. Beasley, J.V. Carcello, D.R. Hermanson, Fraudulent Financial Reporting [17] G.W. Oehlert, Design and Analysis of Experiments: Response Surface Design, W.H.
1987–1991: An Analysis of U.S. Public Companies, The CoSo Report, 1999. Freeman and Company, New York, 2000.
[3] T.B. Bell, J.V. Carcello, A Decision aid for assessing the likelihood of fraudulent [18] O. Persons, Using financial statement data to identify factors associated with
financial reporting, Auditing 19 (1) (2000) 169–184. fraudulent financing reporting, Journal of Applied Business Research 11 (1995)
[4] R.A. Bernardi, Fraud detection: the effect of client integrity and competence and 38–46.
auditor cognitive style, Auditing: A Journal of Practice & Theory, Supplement 13 [19] Z. Rezaee, Financial Statement Fraud-Prevention and Detection, John Wiley &
(1994) 68–84. Sons, Inc., 2002.
[5] R.J. Bolton, D.J. Hand, Statistical fraud detection: A review, Statistical Science 17 [20] H.M. Schilit, Financial Shenanigans, McGraw-Hill, New York, NY, 2002.
(3) (2002) 235–249. [21] C. Spathis, Detecting false financial statements using published. data: some
[6] W.-S. Chen, Y.-K. Du, Using neural networks and data mining techniques for the evidence from Greece, Managerial Auditing Journal 17 (4) (2002) 179–191.
financial distress prediction model, Expert Systems with Applications Part 2 36 [22] C. Spathis, M. Doumpos, C. Zopounidis, Detecting falsified financial statements: a
(2) (March 2009) 4075–4086. comparative study using multicriteria analysis and multivariate statistical
[7] Deloitte, Ten things about financial statement fraud: a review of SEC enforcement techniques, European Accounting Review 11 (3) (2002) 509–535.
releases 2000–2006, Deloitte Forensic Center, June 2007. [23] N.S.R.T.H., W.J. Roux, Response surface approximations for structural optimization,
[8] Y. Dianmin, W. Xiaodan, W. Yunfeng, L. Yue, C. Chao-Hsien, A review of data 1998, pp. 517–534.
mining-based financial fraud detection research, International Conference on [24] H. White, Learning in artificial neural networks: a statistical perspective, Neural
Wireless Communications, Networking and Mobile Computing, 2007, Computation 1 (1989) 425–464.
pp. 5519–5522.
[9] K. Fanning, K.O. Cogger, R. Srivastava, Detection of management fraud: a neural
network approach, 11th Conference on Artificial Intelligence for Applications, Wei Zhou is an Assistant Professor in the Information and
1995, pp. 220–223. Operations Management department at ESCP Europe. His
[10] E.H. Feroz, M.K. Taek, V.S. Pastena, K. Park, The Efficacy of red flags in predicting research interests include Internet advertising, RFID —
the sec's targets: an artificial neural networks approach, International Journal of enabled item-level information visibility, data mining and
Intelligent Systems in Accounting, Finance & Management 9 (2000) 145–157. its application in supply chain management. His work has
[11] J.V. Hansen, J.B. McDonald, W.F. Messier, A generalized qualitative-response appeared in Decision Support Systems, European Journal of
model and the analysis of management fraud, Management Science 42 (1997) Information Systems, European Journal of Operational
1022–1032. Research, IEEE transaction on Geosciences and Remote
[12] E. Kirkos, C. Spathis, Y. Manolopoulos, Data mining techniques for the detection of Sensing, International Journal of Electronic Commerce and
fraudulent financial statements, Expert Systems with Applications 32 (4) (2007) Optical Engineering, among others.
[13] H.C. Koh, C.K. Low, Going concern prediction using data mining techniques,
Managerial Auditing Journal 19 (3) (2004) 462–476.
[14] K.O.C., Kurt M. Fanning, Neural network detection of management fraud using
published financial data, Intelligent Systems in Accounting, Finance & Management Gaurav Kapoor received his Ph.D. in Information Systems from the University of
7 (1) (1998) 21–41. Florida. His research interests include RFID systems.

View publication stats