Scholarlink

To tackle extreme class imbalance, non-independent and identically (non-IID) distributed transactions and dynamic fraud patterns in credit card fraud detection (CCFD), we propose FDFRL, credit card Fraud Detection based on Federated Reinforcement Learning, with three core modules: Kernel-guided adversarial representation learning for high-fidelity synthetic sample generation and robust feature embeddings via hierarchical adversarial refinement. PPO-based federated reinforcement learning employing gradient smoothing and dynamic weighting to harmonize updates from heterogeneous clients and ensure steady convergence. Label-driven federated fusion module that seamlessly integrates local representations into a unified global classifier. Extensive experiments on real-world fraud datasets show that FDFRL markedly outperforms eight state-of-the-art baselines.

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

FDFRL: Credit Card Fraud Detection Based on Federated Reinforcement Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Nana Zhang,
  • Qin Li,
  • Kun Zhu,
  • Dandan Zhu

摘要

To tackle extreme class imbalance, non-independent and identically (non-IID) distributed transactions and dynamic fraud patterns in credit card fraud detection (CCFD), we propose FDFRL, credit card Fraud Detection based on Federated Reinforcement Learning, with three core modules: Kernel-guided adversarial representation learning for high-fidelity synthetic sample generation and robust feature embeddings via hierarchical adversarial refinement. PPO-based federated reinforcement learning employing gradient smoothing and dynamic weighting to harmonize updates from heterogeneous clients and ensure steady convergence. Label-driven federated fusion module that seamlessly integrates local representations into a unified global classifier. Extensive experiments on real-world fraud datasets show that FDFRL markedly outperforms eight state-of-the-art baselines.