WDAE-GAN: A hybrid dual autoencoder and generative adversarial framework with wavelet denoising for credit card fraud detection

Credit card fraud detection remains a critical challenge due to the severe class imbalance in real-world transaction datasets, where fraudulent cases represent only a minute fraction of total records. This study proposes WDAE-GAN, a hybrid detection framework that combines a Wasserstein Generative Adversarial Network (WGAN), dual autoencoders, wavelet-based denoising, and CatBoost classification. In this approach, the WGAN generates realistic synthetic fraud samples to augment the minority class, while dual autoencoders learn distinct latent representations from normal and fraud-augmented data. The concatenated latent features are refined through wavelet denoising before final classification by CatBoost, enhancing feature quality and reducing noise. Experimental results on two benchmark datasets—the European Credit Card Fraud Detection dataset and the IEEE-CIS Fraud Detection dataset—demonstrate that WDAE-GAN achieves near-perfect detection performance. On the European dataset, the model achieved a recall of 0.9999, precision of 0.9999, F1-score of 0.9999, AUC of 1.0000, and AUPRC of 0.9994. On the IEEE-CIS dataset, WDAE-GAN obtained a recall of 0.9994, precision of 0.9994, F1-score of 0.9994, AUC of 1.0000, and AUPRC of 0.9999. These results show that the proposed model not only delivers exceptional detection accuracy but also performs competitively against state-of-the-art methods, effectively identifying rare fraud instances while maintaining an extremely low false-positive rate. This confirms WDAE-GAN’s robustness and scalability for real-world financial fraud detection applications.