A Federated Deep Learning Framework for Privacy-Preserving Digital Identity Anomaly Detection in Distributed Networks

Abstract

The increasing use of distributed digital services has created significant challenges in detecting identity anomalies while protecting sensitive user information. This study develops a federated deep learning framework for privacy-preserving digital identity anomaly detection in distributed networks. The research focuses on decentralized learning, where identity-related data remain within participating client environments, and only model parameters are exchanged for collaborative training. The methodology includes data preprocessing, local deep learning, federated parameter aggregation, privacy protection, anomaly classification, and performance evaluation. The experimental results demonstrate that the proposed approach achieves 98.76% accuracy, 98.42% precision, 98.61% recall, and 98.51% F1-score. The federated training analysis further shows progressive improvement across communication rounds, reaching 97.68% validation accuracy and a 98.21% detection rate. These findings indicate that federated deep learning can effectively support distributed identity anomaly detection while reducing direct exposure of sensitive identity data. The framework provides a foundation for secure and scalable identity analytics in networked environments.