Fraud Detection Model by Using Support Vector Machine Techniques

Abstract

The purpose of this paper is to establish a fraud detection model and explores which of these two data mining methods is the best to detect companies with potential fraud problems. This study uses Support Vector Machine techniques and Logistic Regression to develop fraud detection models combined with five variables: all, filtered, financial, corporate governance, z-score. The results show that model with filtered variables is the most suitable one to detect financial statement fraud one year before the outburst of fraud event and support vector machine with all variables is the most suitable one to detect financial statement fraud two year advance to the outburst of fraud event. The evidence provided by this study minimizes the increasing risks faced by public investors, creditors or external auditors. These people desperately need an effective tool to analyze the disclosed information to make decision. The cost of fraud, both financially and to an organization’s reputation, can be reduced by detecting the fraud in advance. This paper assists management to access the organization’s fraud risk. Our results also offer policymakers with the ability to evaluate the policy implications of corporate governance mechanisms as well as to formulate future policies.