FedAnomDetect a Clustered Federated Learning Framework for Robust Anomaly Detection With Enhanced Privacy and Communication Efficiency

Abstract

Anomaly detection in distributed environments poses significant challenges, particularly in balancing privacy, communication overhead, and detection accuracy. This paper presents FedAnomDetect, a novel federated learning (FL-based framework designed for anomaly detection across large-scale, distributed systems. FedAnomDetect incorporates advanced privacy-preserving techniques, such as differential privacy, to ensure data security during the model aggregation process, while introducing adversarial robustness mechanisms to defend against model poisoning and adversarial perturbations. The framework is further optimized to reduce communication overhead through model compression and clustered FL, ensuring scalability across networks with thousands of edge devices. Experimental results demonstrate that FedAnomDetect outperforms traditional anomaly detection approaches in terms of detection accuracy, false positive rate, and communication efficiency. Additionally, we present a detailed analysis of the framework’s resilience under adversarial conditions and provide quantitative results on latency and scalability in real-world deployments. These findings validate the feasibility of FedAnomDetect for anomaly detection in large-scale distributed networks while maintaining high privacy standards and minimizing computational costs.