Federated Learning for Cybersecurity A Privacy-Preserving Approach

The increasing frequency of cyber threats and the enforcement of strict privacy regulations have exposed critical limitations of traditional centralized machine learning models, especially in distributed environments such as the Internet of Things (IoT). This study presents a federated learning (FL) framework tailored for intrusion detection and malware classification that enables decentralized model training while preserving data locality and minimizing communication overhead. The proposed architecture incorporates lightweight privacy-preserving techniques-including gradient clipping, differential privacy, and encrypted model aggregation-to ensure secure and efficient collaboration across heterogeneous clients. Experimental results on benchmark datasets, such as CICIDS2017 [1] and TON_IoT [2], show that the framework achieves detection accuracies above 90%, while maintaining privacy loss below 5% and improving communication efficiency by more than 25%. These results confirm the viability of federated learning as a scalable and privacy-compliant approach for next-generation cybersecurity systems in highly distributed infrastructures.