Utilizing GANs for Credit Card Fraud Detection: A Comparison of Supervised Learning Algorithms

Abstract

The evolution and improvements in electronic commerce and communications around the world have stimulated credit card use. With the support of smartphone wallets, electronic payments have become the most popular payment method for personal and business use; however, the past few years have also seen a major increase in fraudulent transactions. Corporations and individuals experience very negative impacts from such fraud. Therefore, fraud detection systems have received a lot of attention recently from major financial institutions. This paper proposes a fraud detection approach that deals with small and imbalanced datasets using Generative Adversarial Networks (GANs) for sample generation. Six machine-learning algorithms were applied to real-world data. The accuracy of all six algorithms was above 85% and the precision was above 95%. Five of the six algorithms had a recall score greater than 90%. Furthermore, the Receiver Operating Characteristics (ROC), which measure performance at different thresholds, demonstrated scores greater than 0.90, except Naïve Bayes, which scored 0.81. The proposed approach outperformed the same algorithms in other studies.