Ensemble Method for Credit Card Fraud Detection

Abstract

Traditional credit card fraud detection methods are based on expert experience, while autonomous and intelligent fraud detection has become a direction of study and application. The main problems in fraud detection are class imbalance, concept drift and verification latency. Many works have been proposed to solve class imbalance and concept drift, but few of them consider verification latency. In this paper, we focus on this problem and propose an autonomous and intelligent new method called FD for credit card fraud detection. First, we propose a noise-robust boosting method called SCBoost. Second, we propose a method called k-SDR based on the idea of clustering, it regards a small amount of labeled data as cluster center and conducts clustering according to the distance ratio. Finally, our FD integrates SCBoost and k-SDR. Experiments are conducted to verify the effectiveness of SCBoost, k-SDR and FD.