Novel questionnaire-responded transaction approach with SVM for credit card fraud detection

Abstract

One of the most potential methods to prevent credit card fraud is the questionnaire-responded transaction (QRT) approach. Unlike traditional approaches founded on past real transaction data, the QRT approach proposes to develop a personalized model to avoid credit card frauds from the initial use of new cards. Though this approach is promising, there are still some issues needed investigating. One of the most important issues concerning the QRT approach is how to predict accurately with only few data. The purpose of this paper is to investigate the prediction accuracy of this approach by using support vector machines (SVMs). Over-sampling, majority voting, and hierarchical SVMs are employed to investigate their influences on the prediction accuracy. Our results show that the QRT approach is effective in obtaining high prediction accuracy. They also show that combined strategies, such as weighting and voting, majority voting, and hierarchical SVMs can increase detection rate considerably. © Springer-Verlag Berlin Heidelberg 2005.