A Noise-Resistant Model for Graph-based Fraud Detection

Zhengyang Liu, Hang Yu, Xiangfeng Luo

Abstract: Highlights•Key Contributions:∘<math><mo is="true">∘</mo></math> analyze how noisy labels hurt GNN-based fraud detection via label and neighbor errors∘<math><mo is="true">∘</mo></math> propose a dual-filter GNN with edge-aware aggregation and CVAE-based augmentation∘<math><mo is="true">∘</mo></math> CVAE consistency loss boosts training by assessing low-confidence label quality•Methodology:∘<math><mo is="true">∘</mo></math> propose a noise-resistant GNN with CVAE-based augmentation for robust fraud detection•Application:∘<math><mo is="true">∘</mo></math> applicable to fraud detection in graph data with noisy annotations•Comparative Advantage:∘<math><mo is="true">∘</mo></math> our approach outperforms state-of-the-art methods in noisy label handling