Credit Card Fraud Detection via Integrated Account and Transaction Submodules

摘要

Globally, credit card fraud is a prevalent dilemma. Credit card fraud detection is a classification problem where one aim isto classify legitimate and fraudulent transactions in an adaptive and an automated manner. This paper proposes to utilize anovel hybrid scheme that integrates two mechanisms: a universal model and a unique model. The universal model is a staticmechanism that inspects transactions without regard to the cardholder’s history or any other related transaction. It does soby implementing rules that are obtained via analyzing the complete population. On the other hand, the unique model is adynamic, behavioral scheme that establishes a separate profile for each respective cardholder. In doing so, the model canestablish a specific and accurate system that judges said cardholder’s transactions. It was found that the integration of thetwo models greatly enhanced the performance of the overall system. The system is inherently capable of handling the classimbalance problem that is usually prevalent in credit card fraud classification. The proposed framework was implementedand tested on a typical dataset. The proposed framework exhibited superior performance when benchmarked with similarframeworks. It showed a very high fraud detection rate, high balanced classification rate, high Matthews’ correlation coefficientand a very minimal false alarm rate.