GAN-based Data Augmentation for Credit Card Fraud Detection

Abstract

Deep generative approaches, such as GANs (generative adversarial networks), can be used to efficiently generate new data points that are similar to existing ones. This can be useful for increasing the size of a dataset or for creating synthetic data points that can be used in place of real ones. In this study, we trained classifiers using our novel K-CGAN approach and compared them to other oversampling approaches. We achieved higher F1 score performance metrics than the other methods. After conducting several experiments, we found that classifiers based on a Random Forest, Nearest Neighbor, Logistic Regression, MLP or Adaboost algorithm trained on the augmented set performed much better than those trained on the original data. This effectively creates a fraud detection mechanism.