Feature reduction in federated learning for intrusion detection in IoT networks

Abstract The rapid growth of the Internet of Things (IoT) has significantly increased the complexity of device interactions, making IoT networks more vulnerable to sophisticated cyber threats. Effective intrusion detection is therefore crucial to ensuring the security and resilience of these systems. This paper presents federated learning with feature reduction (Fed-FeRe), a novel approach that enhances decentralized intrusion detection by integrating $$\chi ^{2}$$ χ 2 -based feature selection with a gated recurrent unit model. Fed-FeRe introduces an adaptive initialization of the performance threshold $$\alpha $$ α and a data-driven estimation of key hyperparameters ( $$\theta _{0}$$ θ 0 , $$\eta $$ η ), enabling robust performance across diverse IoT conditions. By dynamically optimizing feature selection, the framework reduces computational overhead and communication costs, achieving approximately 2% lower transmission costs than FedAvg and 17.6% lower GPU utilization than MOON (Model-Contrastive Federated Learning), while improving detection precision by 3.06%. Fed-FeRe further demonstrates scalability to varying client sizes and adaptability to distinct IoT application domains such as smart cities, healthcare, and industrial networks. These results highlight Fed-FeRe as a scalable, efficient, and privacy-preserving solution for real-world IoT security, advancing the state of federated learning-based intrusion detection.