Implementation on Credit Card Fraud Detection using GAN Oversampling

In recent years, online payments through credit cards and UPI have seen a significant surge in transactions. A significant problem for financial institutions is the rise in fraudulent activity that has coincided with the increased dependence on card payments. Fraudsters employ various methods to obtain card information illegally, such as stealing physical cards, card swapping, phishing, or data breaches. With the majority of transactions being valid and only a small percentage being fraudulent, transaction data in this context is frequently extremely unbalanced. This disparity in class may result in erroneous metric evaluation, poor generalization, and model bias. By creating artificial instances of the minority class, Generative Adversarial Networks (GAN) have been used to balance the dataset in order to solve this problem. The Synthetic Minority Oversampling Technique (SMOTE), which is compared to GANs have demonstrated superior performance, achieving higher F1 scores in fraud detection models. High dimensionality and highly imbalanced datasets are especially well-suited for GAN processing. With its notable class imbalance, the European Cardholders 2013 dataset is used in this study to detect credit card fraud transactions. The study emphasizes how GANs can effectively detect fraudulent transactions and improve predictive model performance.