A Time-Aware Generative Network for Enhancing Transaction Security in Consumer Electronics

As consumer electronics increasingly rely on secure digital transactions, ensuring transaction security remains a formidable challenge, largely due to the severe class imbalance between legitimate and fraudulent transactions. Although existing methods attempt to mitigate this imbalance, their effectiveness is often compromised by the sophisticated concealment strategies employed by fraudsters, which lead to significant behavioral overlap between legitimate and fraudulent transactions. In this paper, we propose a novel Time-aware Generative Network (TaGN) that combines a Density-based Wasserstein Generative Adversarial Network (DWGAN) with a Time-aware Gated Recurrent Network (TaGRN) to enhance fraud detection in consumer electronics. The DWGAN penalizes synthetic samples that closely resemble legitimate transactions, focusing the generator on fraudulent characteristics and reducing the impact of behavioral overlap. The TaGRN leverages temporal and contextual transactional information through a gated recurrent structure, transforming original transaction features into new representations that effectively distinguish behavior patterns in overlapping regions. Extensive experiments conducted on real-world and public datasets demonstrate that TaGN significantly outperforms its competitive peers, offering a robust and effective framework for enhancing transaction security in consumer electronics.