A Machine Learning-Based Energy-Efficient Clustering Framework for Dynamic IoT Networks

The Internet of Things and Wireless Sensor Networks have become indispensable for a variety of applications. Scaling these networks requires efficient clustering and cluster head selection in order to manage energy consumption and maintain network performance. This paper proposes a novel clustering framework called Federated Learning for Head Selection and Spectral Clustering (FLHC-SC), which uniquely combines Federated Learning for decentralized cluster head (CH) selection and spectral clustering for graph-based clustering. Unlike conventional methods in which clustering is performed before CH selection, FLHC-SC reverses this process, making it possible to select CHs intelligently based on local learning prior to spatial partitioning. As a result of this decoupled, two-phase strategy, FLHC-SC can adapt to dynamic network conditions while maintaining energy and structural efficiency. IoT simulations were conducted to compare the proposed method with three recently proposed clustering methods: a modified version of Low-Energy Adaptive Clustering Hierarchy (LEACH Distributed Clustering), a hybrid PSO-based approach and a K-means–fuzzy logic clustering method. Results show that FLHC-SC outperforms other methods in several key performance metrics. It maintains 60% of active nodes at the end of simulation rounds while preserving a higher average residual energy. Furthermore, it achieves a maximum silhouette coefficient of 0.998 and a lower intra-cluster compactness value. Statistical validation with ANOVA test was used to confirm the superior performance of the proposed method.