Scholarlink

<p>The exponential growth of e-commerce and digital payment systems has brought significant convenience to users. Meanwhile, it has also led to a surge in fraudulent activities in online transactions. Existing fraud detection mechanisms often lack critical features like real-time alerts, transaction history, dynamic updates, fraud scoring, and continuous monitoring, all of which are vital for building an effective fraud detection system. This work proposes <i>Fedshield</i>, a decentralized, federated learning-based fraud detection system that ensures trust among participants while preserving data privacy in fintech environments. <i>Fedshield</i> used open source high performance computing (OpenMPI) library to manage model synchronization and real-time fraud detection. To handle dynamic shifts in transaction patterns, we introduce a moving time frame approach that keeps the risk prediction model resilient to evolving fraud tactics. Experimental results demonstrate that <i>Fedshield</i> surpasses existing state-of-the-art methods, achieving an impressive F1-score of 0.9903 using newly engineered features. A dashboard is designed with Django and integrated via a Flask API, enabling efficient data flow management, while SQLite is used to store transaction history. This dashboard supports real-time monitoring of suspicious activities and provides instant alerts to both users and issuing banks.</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

FedShield: federated learning based robust online payment fraud detection

  • 期刊论文

  • 发表日期：

    2025年8月29日

查看全文
  • Narendra Singh,
  • Somanath Tripathy

摘要

The exponential growth of e-commerce and digital payment systems has brought significant convenience to users. Meanwhile, it has also led to a surge in fraudulent activities in online transactions. Existing fraud detection mechanisms often lack critical features like real-time alerts, transaction history, dynamic updates, fraud scoring, and continuous monitoring, all of which are vital for building an effective fraud detection system. This work proposes Fedshield, a decentralized, federated learning-based fraud detection system that ensures trust among participants while preserving data privacy in fintech environments. Fedshield used open source high performance computing (OpenMPI) library to manage model synchronization and real-time fraud detection. To handle dynamic shifts in transaction patterns, we introduce a moving time frame approach that keeps the risk prediction model resilient to evolving fraud tactics. Experimental results demonstrate that Fedshield surpasses existing state-of-the-art methods, achieving an impressive F1-score of 0.9903 using newly engineered features. A dashboard is designed with Django and integrated via a Flask API, enabling efficient data flow management, while SQLite is used to store transaction history. This dashboard supports real-time monitoring of suspicious activities and provides instant alerts to both users and issuing banks.