A Hybrid Profiling Method to Detect Heterogeneous Credit Card Frauds

The continued growth of online shopping, which is naturally followed by an increase in associated online frauds, exposes merchants to potentially huge financial losses. Hence many researches have been dedicated to detecting online payment frauds. Addressing the heterogeneous behavior of fraudsters which leads to different types of frauds has always been a challenge. In this paper, a hybrid profiling system based on anomaly detection technique for credit card fraud detection which combines transaction-level and aggregation-level profiles has been proposed. In the proposed approach due to an observed seasonal behavior of cardholders, a two-level clustering method is used to construct a transaction-level profile of each cardholder. This stage groups similar monthly behaviors of cardholders in the first level and clusters transactions of each group separately in the next level. To construct the aggregation-level profile, a method for transaction aggregation is applied as another strategy for fraud detection and the online and offline usage of aggregated data is considered in the proposed system. Results indicate that the parts of the system are complementary and combining them has compensated for their individual deficiencies, improved detection rate, resulted in a timelier fraud detection, and consequently more monetary saving. Furthermore, we demonstrate that the two-level clustering has reduced false alarms significantly

Keywords:

Fraud Detection , Clustering , Profile , Aggregation , Credit Card , Hybrid System