The landscape of credit card fraud is evolving rapidly, with the emergence of increasingly sophisticated fraudulent methods. This trend has resulted in a significant uptick in financial losses incurred by both businesses and consumers. To address the challenge of credit card fraud detection, the industry has widely adopted machine learning models. However, building effective models is hindered by limited real-world data and the severe imbalance between fraudulent and legitimate transactions. In this work, I explore the application of Generative Adversarial Networks (GANs) to synthesize fraudulent samples and their superiority compared with traditional data augmentation techniques. To mitigate potential biases introduced by a single modeling method, Logistic Regression (LR), Random Forest (RF), and Extreme Gradient Boosting (XGB) are employed, representing different modeling paradigms. My experiments show that models trained on the GANs-based synthetic data exhibit superior generalization capabilities and a stronger ability to discriminate between different classes.
