Optimization of Financial Fraud Detection Algorithm by Combining Variational Autoencoder with Generative Adversarial Networks

This article delves into the field of financial fraud detection and proposes innovative solutions based on big data, machine learning, and deep learning technologies to address the challenges posed by the complexity and technology driven new features of modern financial fraud that traditional credit evaluation methods face. The latest progress in financial fraud risk, detection technology, and machine learning applications was systematically reviewed through literature research, and the shortcomings of existing research were identified. To solve the problem of sample imbalance, this paper innovatively designs AE-G and VAE-G models, which use normal transaction samples to train and generate fraudulent samples, achieving efficient binary classification learning and effectively alleviating the problem of data imbalance. In empirical analysis, the excellent performance of AE-G and VAE-G models in fraud detection was verified using the credit card transaction dataset jointly developed by the Free University of Brussels and Worldline. In particular, the VAE-G model performed well in AUC, recall rate, and F1 score, and significantly improved the recognition ability of minority fraud transactions. This article not only provides new ideas for the field of financial fraud detection, but also provides technical support for practical applications. At the same time, it points out the limitations of dataset diversity and the authenticity of generated fraud samples, which provides direction for future research and is of great significance for promoting the development of financial technology in the field of anti fraud.