ABSTRACT The widespread adoption and deployment of the Internet of Things (IoT) have significantly transformed modern digital ecosystems. However, this expansion has also introduced new security vulnerabilities, making real‐time and efficient intrusion detection an increasingly important research area. Traditional intrusion detection systems often rely on centralized data aggregation, which raises challenges related to scalability, privacy, and latency. In this context, developing distributed intrusion detection systems that preserve the privacy of continuously streaming IoT data remains less explored problem. Addressing this gap is essential to ensure secure and adaptive monitoring within dynamic IoT environments. This paper proposes an intrusion detection framework that combines Hoeffding Tree‐based incremental learning with federated learning enhanced by using generalized entropy measures, specifically Renyi and Tsallis. The Renyi entropy provides a mechanism to tune the sensitivity of detection through an adjustable order parameter. Tsallis entropy effectively models non‐extensive behaviors in data, which is particularly useful for identifying rare and anomalous events such as cyber attacks. A new algorithm is introduced to estimate the entropic parameters of Renyi and Tsallis entropies directly from data by maximizing information gain. The proposed federated Hoeffding Tree variations are evaluated against a baseline model using standard IoT benchmark datasets. Statistical validation demonstrates that the generalized Renyi and Tsallis entropy‐based models outperform the baseline, achieving up to a 9% improvement in both accuracy and F1‐score. These findings provide valuable insights for designing scalable, privacy‐preserving, and robust intrusion detection systems suited for next‐generation IoT security frameworks.
Dey et al. (Fri,) studied this question.