Scholarlink

<p>The detection of fraudulent transactions remains a critical challenge in the financial services industry, further intensified by the rapid growth in transaction volumes and the increasingly sophisticated tactics of fraudsters. Traditional methods, including current graph neural networks, have shown potential but often fall short in accurately detecting fraud due to the ability of fraudsters to mimic legitimate behaviors, conceal relationships, and operate within narrow time frames. These limitations necessitate the development of more effective detection techniques. This study is motivated by the need to overcome the inherent limitations of existing GNN-based methods in fraud detection. We propose an innovative approach that enhances local information aggregation and incorporates global insights through graph reconstruction, addressing both the structural and temporal complexities introduced by fraudulent activities. Our method utilizes attention mechanisms to capture spatial–temporal relationships within transaction networks, emphasizing crucial connections between nodes to improve detection efficiency. Key innovations include attentional intra-relation aggregation, which differentiates the strength of connections between neighboring nodes, and attentional inter-relation aggregation, which prioritizes nodes that share a higher number of common neighbors. We validate the effectiveness of the proposed method using two benchmark datasets, YelpChi and Amazon, as well as a real-world dataset. Experimental results demonstrate that our approach significantly outperforms existing techniques, achieving notable improvements in AUC, F1-macro, and Gmean metrics. These results highlight the importance of addressing the current limitations in GNN-based methods and demonstrate the potential of our approach to provide a more robust solution for financial fraud detection.</p>

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

Transaction fraud detection via attentional spatial–temporal GNN

  • 期刊论文

  • 发表日期：

    2025年2月24日

查看全文
  • Samiyeh Khosravi,
  • Mehrdad Kargari,
  • Babak Teimourpour,
  • Mohammad Talebi

摘要

The detection of fraudulent transactions remains a critical challenge in the financial services industry, further intensified by the rapid growth in transaction volumes and the increasingly sophisticated tactics of fraudsters. Traditional methods, including current graph neural networks, have shown potential but often fall short in accurately detecting fraud due to the ability of fraudsters to mimic legitimate behaviors, conceal relationships, and operate within narrow time frames. These limitations necessitate the development of more effective detection techniques. This study is motivated by the need to overcome the inherent limitations of existing GNN-based methods in fraud detection. We propose an innovative approach that enhances local information aggregation and incorporates global insights through graph reconstruction, addressing both the structural and temporal complexities introduced by fraudulent activities. Our method utilizes attention mechanisms to capture spatial–temporal relationships within transaction networks, emphasizing crucial connections between nodes to improve detection efficiency. Key innovations include attentional intra-relation aggregation, which differentiates the strength of connections between neighboring nodes, and attentional inter-relation aggregation, which prioritizes nodes that share a higher number of common neighbors. We validate the effectiveness of the proposed method using two benchmark datasets, YelpChi and Amazon, as well as a real-world dataset. Experimental results demonstrate that our approach significantly outperforms existing techniques, achieving notable improvements in AUC, F1-macro, and Gmean metrics. These results highlight the importance of addressing the current limitations in GNN-based methods and demonstrate the potential of our approach to provide a more robust solution for financial fraud detection.