Effective Data Generation for E-banking Transactions Using Cycle-Consistent Adversarial Networks

Abstract

In the anti-fraud research, the small amount of fraudulent transactions leads to extremely class imbalanced data. This problem becomes a bottleneck of fraud detection. In this paper, we apply the Cycle-Consistent Adversarial Networks (CycleGAN) to generate data for the minority class. Based on real e-banking transaction data, generators and discriminators are designed to generate synthetic data that meets the characteristics of e-banking transactions. Synthetic samples and real samples are mixed into the training of fraud detection model, and multiple metrics are used to verify the effect. Experimental results show that the synthetic data generated by CycleGAN can effectively improve the performance of the fraud detection model.