A Feature Extraction Method for Credit Card Fraud Detection

As credit card fraud has caused huge economic losses and harm cardholders seriously, credit card fraud detection is important and has been paid much attention. An effective feature engineering is the key to build an effective model of detecting fraud. However, we find that the current feature engineering that is based on the frequency of transactions is not perfect. Although frequency-based feature engineering depicts the temporal features of user transactions, it does not enough consider the fraud characteristics and the distinction of transaction behaviors. Starting from the two points, we propose a rule-based feature engineering that considers both individual behavior and group behavior, and portrays the individual behavior as group features, and thus can more effectively distinguish legitimate and fraudulent transactions. Our experiments illustrate the advantages of our method.