A Behavior-cluster Based Imbalanced Classification Method for Credit Card Fraud Detection

Credit card fraud detection has been paid more and more attention by researchers. The credit card transactions are represented by highly imbalanced data sets. The number of genuine transactions is far more than fraudulent transactions, which will greatly affect the detection of fraud. Existing methods mainly consider how to balance the two classes only based on data volume, without considering the complexity of user behavior in credit card transactions, that is, the behavior noise. In this paper, we propose a behavior-cluster based imbalanced classification method. The main idea is to divide user behaviors into several group behaviors, remove behavior noise, and then hierarchical sampling. Experiments on a large scale credit card transaction data provided by a financial institution and 18 UCI data sets show that our method is superior to the existing method.