AD-VGF: An Improved Generative Adversarial Network for Credit Card Risk Identification and Management in Digital Economy
Yan Zhang, Zhang Yuqi, Yongtang Liang, Shengdong Mu, Zhuoming Zhang, Xiao Jie Chen · Cybernetics & Systems · 2025
Credit card fraud faces the problem of imbalanced data categories, in this paper, we consider the minority fraud data as outliers and recognize them directly, and generate an improved generative adversarial model AD-VGF (Anomaly Detection Variational Gradient Flow) through variational gradient flow learning on probabilistic measurement space for handle highly unbalanced datasets. The validity of the model is verified using the Kaggle Group dataset, and the average AUC of the AD-VGF model reaches 95.5% with the increase of Epoch, which is about 10% improvement compared with the performance of the traditional models. The AD-VGF model is able to identify very few fraudulent samples in the large samples directly, which makes the identification more accurate and efficient.