Suspicious Behavior Detection in Debit Card Transactions using Data Mining: A Comparative Study using Hybrid Models

Suspicious Behavior Detection in Debit Card Transactions using Data Mining: A Comparative Study using Hybrid Models

Ehsan Saghehei (Department of Industrial Engineering, Islamic Azad University Malayer Branch, Malayer, Iran) and Azizollah Memariani (Department of Mathematics and Computer Science, University of Economic Sciences, Tehran, Iran)
Copyright: © 2015 |Pages: 14
DOI: 10.4018/IRMJ.2015070101
OnDemand PDF Download:
$37.50

Abstract

The approach used in this paper is an implementation of a data mining process against real-life transactions of debit cards with the aim of detecting suspicious behavior. The framework designed for this purpose has been obtained through merging supervised and unsupervised models. First, due to unlabeled data, Twostep and Self-Organizing Map algorithms have been used in clustering the transactions. A C5.0 classification algorithm has been applied to evaluate supervised models and also to detect suspicious behaviors. An innovative plan has been designed to evaluate hybrid models and select the most appropriate model for the solution of the fraud detection problem. The evaluation of the models and the final analysis of the data took place in four stages. The appropriate hybrid model was selected from among 16 models. The results show a high ability of selected model in detecting suspicious behavior in transactions involving debit cards.
Article Preview

1. Introduction

Plastic cards are indispensable parts of the modern payment system. They provide users with a wide variety of services (Krivko, 2010). While these systems enjoy support from advanced software and hardware, the payment process still suffers from inadequate security (Panigrahi, Kundu, Sural, & Majumdar, 2009). This situation has made abuse and fraud possible and has imposed severe harm and much aggravation to the companies issuing the cards. To grasp the importance of the issue, statistics on the quantity of plastic card frauds must be reviewed. Actually, determining the amount of plastic card fraud is not an easy task, in part due to the companies’ reluctance to reveal fraud numbers publicly, partly to temper their customers’ dismay, and in part due to the volatility of fraud numbers reported during any one time period. Nevertheless, numerous estimates exist. For example, Aleskerov et al. cited estimates of $700 million in the United States each year for Visa/Mastercard fraud and $10 billion worldwide in 1996 (Bolton & Hand, 2002). In 2013, total amount of card fraud losses for cards issued in UK was reported to be £450.4 million (APACS, 2014).

There are various types of fraud in the plastic card industry. For a detailed classification one can refer to Kou, Lu, Sirwongwattana, & Huang (2004). As better efforts are made to detect and defeat fraud, the perpetrators become more devious and sophisticated, and the whole system becomes more complex. There is a need for continuous innovation in approaches to fraud prevention, detection and control. Financial fraud detection methods have been divided into two broad categories: supervised and unsupervised (Bolton & Hand, 2001). Phua et al.(2005) categorized fraud detection techniques in four groups: supervised, hybrid (supervised and unsupervised), semi supervised and unsupervised. In general hybrid models are comprised of two or more techniques of either supervised or unsupervised or combination of both.

In the supervised group, artificial neural network techniques have been used frequently (Aleskerov, Freisleben, & Rao, 1997; Dorronsoro, Ginel, Sgnchez, & Cruz, 1997; Ghosh & Reilly, 1994; Maes, Tuyls, Vanschoenwinkel, & Manderick, 2002; Syeda, Zhang, & Pan, 2002). Also recently Akhilomen (2013) used this technique for cyber credit card fraud detection. Another widely used technique in the field of supervised models is the decision tree approach. Examples are Wang, Fan, Yu, & Han (2003) and Wei (2004). Bhattacharyya et al. (2011) applied support vector machines and random forests techniques as a comparative study. Their article was based on real-life data obtained from transactions from an international credit card operation.

Complete Article List

Search this Journal:
Reset
Open Access Articles: Forthcoming
Volume 30: 4 Issues (2017)
Volume 29: 4 Issues (2016)
Volume 28: 4 Issues (2015)
Volume 27: 4 Issues (2014)
Volume 26: 4 Issues (2013)
Volume 25: 4 Issues (2012)
Volume 24: 4 Issues (2011)
Volume 23: 4 Issues (2010)
Volume 22: 4 Issues (2009)
Volume 21: 4 Issues (2008)
Volume 20: 4 Issues (2007)
Volume 19: 4 Issues (2006)
Volume 18: 4 Issues (2005)
Volume 17: 4 Issues (2004)
Volume 16: 4 Issues (2003)
Volume 15: 4 Issues (2002)
Volume 14: 4 Issues (2001)
Volume 13: 4 Issues (2000)
Volume 12: 4 Issues (1999)
Volume 11: 4 Issues (1998)
Volume 10: 4 Issues (1997)
Volume 9: 4 Issues (1996)
Volume 8: 4 Issues (1995)
Volume 7: 4 Issues (1994)
Volume 6: 4 Issues (1993)
Volume 5: 4 Issues (1992)
Volume 4: 4 Issues (1991)
Volume 3: 4 Issues (1990)
Volume 2: 4 Issues (1989)
Volume 1: 1 Issue (1988)
View Complete Journal Contents Listing