A Hybrid Approach for Fraud Detection in Digital Wallet Transactions Using Adversarial Autoencoders and Gated Recurrent Units
J. Shaik, Chaman Kumar, Vedula Sitharamulu, B. N. V. Madhu Babu, Hanumantha Rao Battu, Katta Venkataramana · Engineering Technology & Applied Science Research · 2025
Digital payment systems are increasingly used to complete everyday transactions; however, their digital nature exposes users to the risk of fraudulent activity, necessitating advanced detection techniques to ensure security. This study proposes a hybrid fraud detection model for digital wallet transactions by integrating Adversarial Autoencoders (AAE) and Gated Recurrent Units (GRU), combining AAE's ability to learns robust latent representations, and GRU's ability to capture temporal dependencies within transaction sequences. The proposed method outperforms existing approaches, achieving 99% accuracy, 99% recall, 98% F1-score, 99% precision, and 99.4% Area Under the Curve (AUC). By effectively reducing both false positives and false negatives, the model improves fraud detection and mitigates financial risks in digital transactions.