Privacy-Preserving Cyber Threat Detection in Distributed IoT Networks

Abstract

This study aims to examine the applicability of federated learning as a privacy-preserving approach for cyber-threat detection in distributed Internet of Things (IoT) networks. The study addresses the need for collaborative security analysis without requiring participating clients to transfer their raw network data to a centralized repository. The study examines the reported effectiveness of federated learning across distributed IoT clients and the influence of different model aggregation strategies on threat-detection performance. It also considers the performance relationship between federated and centralized learning approaches. The analyzed findings indicate that federated learning can provide consistent cyber-threat detection across participating clients. The reported results show that the aggregated federated model provides competitive classification performance compared with centralized learning. Analysis of the reported results for FedAvg, FedAvgM, FedAdam, and FedAdagrad further indicates that the aggregation strategy affects the performance of the resulting global model. The analyzed results indicate that federated learning can maintain effective detection capability despite the decentralized distribution of training data. The study concludes that federated learning represents a promising approach for privacy-aware cybersecurity in distributed IoT environments. By retaining training data at participating clients and exchanging model updates, the approach can support collaborative threat detection while reducing direct exposure of raw network information. Future research should investigate real-world heterogeneous IoT environments, stronger privacy mechanisms, communication efficiency, and explainable federated security models.