A Hybrid Federated Reinforcement Learning and Game-theoretic Framework for Energy-aware Clustering and Fault Diagnosis in IoT-enabled Wireless Sensor Networks
Year: 2026
Type: article
Abstract: Wireless Sensor Networks (WSNs) are essential for IoT applications, yet conventional clustering protocols suffer from suboptimal cluster-head selection, ignoring node cooperation, privacy, and fault diagnosis.This study proposes a hybrid framework integrating Federated Learning (FL), Reinforcement Learning (RL), and Game Theory (GT) for cluster-head selection and fault diagnosis.The results show that FL, RL and GT are suitable for an effective, scalable and communication-efficient clustering solution for IoT-enabled WSNs.The energy saving results show that the energy saving can reach 9.1% with it compared with the best baseline (FL-DQN-DP-LEACH), 19.0% in Half Node Death.A 97.2% accuracy in fault detection and a 94.8% accuracy in classification with an Area Under the Receiver Operating Characteristic Curve (AUC of the ROC curve) of 0.97.Ablation results show that the performance would be reduced by removing FL, RL and GT with an overall energy consumption of the network increased by 20.3%, 16.3%, and 11.3% respectively from the complete proposed architecture.Statistically significance differences in performance are reported through deployment-level p aired comparisons and exact p-values are reported according to the adjusted significance thresholds.The study ensures full reproducibility through public source code and seed documentation.The results show that the integration of FL, RL and GT is a good decentralized clustering solution for WSNs enabled by IoT devices in terms of being effective, scalable and raw-data-local.
Author: Lateef Abd Zaid Qudr
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7235429979
