Application of Data Mining for Supply Chain

Inventory Forecasting
Nenad Stefanovic1, Dusan Stefanovic2, Bozidar Radenkovic3
1
Information Systems Division, Zastava Automobiles , Kragujevac, Serbia,
stefanovic.n@gmail.com
2
Faculty of Science, University of Kragujevac, Serbia, dusan@kg.ac.yu
3
Faculty of Organizational Sciences, University of Belgrade, Serbia,
boza@ieee.org

Abstract

This paper deals with data mining applications for the supply chain inventory
management. It describes the use of business intelligence (BI) tools, coupled with
data warehouse to employ data mining technology to provide accurate and up-to-
date information for better inventory management decisions. The methodology is
designed to provide out-of-stock forecasts at the store/product level. The first phase
of the modelling process consists of clustering stores in the supply chain based
upon aggregate sales patterns. After quality store-cluster models have been
constructed, these clusters are used to more accurately make out-of-stock
predictions at the store/product level using the decision trees and neural network
mining algorithms. The methods for evaluation and accuracy measurement are
described. Also, the specialized front-end BI web portal that offers integrated
reporting, web analytics, personalization, customization and collaboration is
described.

1. Introduction
,Q WRGD\V IDVW-changing and global environment characterised with high level of
uncertainty, partnership and collaboration become the critical factors for the
ultimate success on the market. Thus companies compete as part of bigger supply
chains. A supply chain is a complex system generating a huge amount of
heterogeneous data. Companies need to turn these data into knowledge, to avoid
EHFRPLQJ GDWD ULFK DQG LQIRUPDWLRQ SRRU It is not enough only to know what
happened and what is happening now, but also what will happen in the future and
how/why did something happen.
What is needed is a unified supply chain intelligence framework to collect,
integrate, consolidate all relevant data and to use business intelligence (BI) tools
like data warehousing and data mining, to discover hidden trends and patterns in
large amounts of data and finally to deliver derived knowledge to the business
users via Web portals [1].
As a fastest growing BI component, data mining allows us comb through our data,
notice patterns, devising rules, and making predictions about the future It can be
defined as the analysis of (often large) observational data sets to find unsuspected
relationships and to summarize the data in novel ways [2].
176 Richard Ellis, Tony Allen and Miltos Petridis (Eds)

2. Inventory Forecasting and Data Mining

Inventory control is the activity which organises the availability of items to the
customers. It coordinates the purchasing, manufacturing and distribution functions
to meet marketing needs.
Inventory management is one of the most important segments of supply chain
management. Companies face the common challenge of ensuring adequate
product/item stock levels across a number of inventory points throughout the
supply chain. Additionally, uncertainty of demand, lead time and production
schedule, and also the demand information distortion known as the bullwhip effect
[3], make it even more difficult to plan and manage inventories.
The basis for decision making should be information about customer demand.
Demand information directly influences inventory control, production scheduling,
and distribution plans of individual companies in the supply chain [4]. Decision
making based on local data leads to inaccurate forecasts, excessive inventory, and
less capacity utilization.
Generally, determining the adequate stock levels balances the following competing
costs:
Overstocking costs these include costs for holding the safety stocks, for
occupying additional storage space and transportation.
Costs of lost sales these are costs when the customer wants to buy a
product that is not available at that moment.
Commonly, managers have relied on a combination of ERP, supply chain, and
other specialized software packages, as well as their intuition to forecast inventory.
+RZHYHULQWRGD\VKLJK XQFHUWDLQHQYLURQPHQWDQGODUJH TXDQWLWLHVRIGLVSDUDWH
data demands new approaches for forecasting inventory across the entire chain.
Data mining tools can be used to more accurately predict stock levels for different
products located at various supply chain nodes.
The best way to deal with these competing costs is to use data mining techniques to
ensure that each inventory point (internal warehouse, work-in-process, distribution
centre, retail store) has the optimal stock levels.
Data mining applies algorithms, such as decision trees, clustering, association, time
series, and so on, to a dataset and analyzes its contents. This analysis produces
patterns, which can be explored for valuable information. Depending on the
underlying algorithm, these patterns can be in the form of trees, rules, clusters, or
simply a set of mathematical formulas. The information found in the patterns can
be used for reporting, as a guide to supply chain strategies, and, most importantly,
for prediction.
Data mining can be applied to the following tasks [5]:
Classification
Estimation
Segmentation
Association
Forecasting
Text analysis