Hierarchical Federated Edge Intelligence Architecture for Transaction-Level Fraud Detection in Digital Banking Systems

Due to the rapid growth of digital banking, the number of fraud cases has risen, and it requires intelligent, safe, and real-time detection systems. The current centralized and fixed models have scale, privacy and cross-institutional limitations. The proposed study presents a new architecture called H-FEFTNet (Hierarchical Federated Edge Fraud Transaction Network), a framework that combines federated edge intelligence, explainable AI and adaptive aggregation systems to detect transaction-level fraud. The proposed model is implemented with the help of the Bank Account Fraud Dataset Suite (NeurIPS 2022) to make collaboration privacy-preserving, low-latency, and better model generalization. Experimental analysis is shown to have better results with an accuracy, precision, recall and F1-score of 98%, 97%, 96%, and 96% respectively, and is better than related works in performance and interpretability. The system is very relevant to real time digital banking ecosystems, which fosters transparency, compliance and resiliency in financial fraud prevention.

Ҳали таржима қилинмаган