Financial Fraud Detection System Combining Generative Adversarial Networks and Deep Learning

In this research, we effectively merge generative adversarial networks (GAN) with deep learning methods and implement them in the area of financial fraud detection. After preprocessing and selecting features from public datasets, we developed a robust GAN model and trained the discriminator by combining real data with fake data generated by the generator. Research findings indicate that the deep learning model utilizing GAN outperforms traditional machine learning techniques in terms of detection accuracy, recall rate, and F1 score. Data enhancement with the GAN model produces fake data closely resembling real data, addressing issues with imbalanced financial fraud transaction data and enhancing the model’s detection capabilities and robustness. In addition, the model was comprehensively evaluated through various indicators such as ROC curve and confusion matrix in the experiment, which further verified the effectiveness and superiority of the GAN model. Overall, this study demonstrates the application value and superior performance of this method in financial fraud detection by building and training a GAN model, combined with data enhancement and experimental verification, and provides an effective solution for improving the security of financial transactions.