Scholarlink

With digital transactions, financial fraud has grown at a dramatic rate, with U.S. losses alone exceeding $ 12.5B per year and compromising the trust of global payment systems. Conventional ML methods fall short against complex relational patterns such as fraud rings and money laundering networks, which require graph-aware modeling. This survey offers a unified framework for Graph Neural Networks (GNNs) in financial fraud detection, addressing three research questions systematically. RQ1 demonstrates that GNNs outperform XGBoost with 12–25% AUROC improvement by relational modeling of fraud rings, contextual propagation, and higher-order dependencies across heterogeneous transaction graphs. RQ2 describes production-ready neural networks-based architectures, imbalance mitigation, heterophily handling, and deployment strategies to achieve <100 ms latency at 10K+ TPS with federated learning, which is validated by real-world studies with gains of 25–45% fraud reduction. RQ3 outlines key challenges such as adversarial camouflage, spatiotemporal limitations, billion-edge scalability, regulatory interpretability and future directions, including causal GNNs, continual learning, and tiered explanation frameworks to place GNNs at the forefront of next-generation fraud prevention against Artificial Intelligence (AI)-augmented financial crime.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Fraud Detection Using Graph Neural Networks: A Survey

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Emrah Arslan,
  • Samiyeh Khosravi,
  • Javad Hassannataj Joloudari,
  • Roohallah Alizadehsani,
  • Juan Manuel Gorriz

摘要

With digital transactions, financial fraud has grown at a dramatic rate, with U.S. losses alone exceeding $ 12.5B per year and compromising the trust of global payment systems. Conventional ML methods fall short against complex relational patterns such as fraud rings and money laundering networks, which require graph-aware modeling. This survey offers a unified framework for Graph Neural Networks (GNNs) in financial fraud detection, addressing three research questions systematically. RQ1 demonstrates that GNNs outperform XGBoost with 12–25% AUROC improvement by relational modeling of fraud rings, contextual propagation, and higher-order dependencies across heterogeneous transaction graphs. RQ2 describes production-ready neural networks-based architectures, imbalance mitigation, heterophily handling, and deployment strategies to achieve <100 ms latency at 10K+ TPS with federated learning, which is validated by real-world studies with gains of 25–45% fraud reduction. RQ3 outlines key challenges such as adversarial camouflage, spatiotemporal limitations, billion-edge scalability, regulatory interpretability and future directions, including causal GNNs, continual learning, and tiered explanation frameworks to place GNNs at the forefront of next-generation fraud prevention against Artificial Intelligence (AI)-augmented financial crime.