A Conceptual Real-Time and Privacy-Preserving Architecture for Banking Fraud Detection

Abstract

Nowadays, banking transaction fraud has led to huge losses for individuals and banks, yielding a need for effective and quick fraud detection systems. This study provides a novel conceptual architecture for banking fraud detection, notably tailored to cope with pivotal issues related to swiftly evolving fraud trends, preserving data privacy, as well as the growing need for real-time detection and integrated systems. The proposed architecture leverages Federated Reinforcement Learning (FRL) to enable adaptive, decentralized model training across multiple institutions while maintaining data privacy. The proposed architecture leverages Federated Reinforcement Learning (FRL) for adaptive, decentralized model training, utilizing a hybrid blockchain for secure and transparent interactions while maintaining data privacy. Real-time Edge AI detects fraud with minimal latency for sending rapid signals, whereas Explainable AI (XAI) provides comprehensive predictions to boost trust and conformity to laws. Ultimately, Dynamic Model Personalization by meta-learning adjusts global models to local fraud habits, improving detection accuracy. Together, these strategies shape a scalable, privacy-preserving architecture for digital banking fraud detection.