Federated Learning Multi-sensor Data Fusion for Privacy-preserving and Wi-Fi Spoofing Detection in Sensor Networks

The rapid expansion of IoT devices in smart campuses has increased vulnerabilities to Wi-Fi spoofing and signal mimicry attacks.Therefore, a decentralized, lightweight physical-layer anomaly detection system is developed in this study, operating autonomously at resourceconstrained edge nodes.The developed federated learning multi-sensor fusion (FL-MSF) system integrates received signal strength, multipath delay spreads, and background noise into a regional fluctuation index (RFI).RFI enables edge transceivers to isolate malicious signal injections from benign fluctuations without relying on heavy centralized classifiers.To address non-independent and identically distributed data across heterogeneous environments, FL-MSF introduces an adaptive aggregation mechanism that weights local model updates based on material attenuation profiles.Model updates are securely combined using the federated averaging (FedAvg) protocol with clipping-based differential privacy, ensuring strong data sovereignty.Simulation results in a high-density campus environment (50 edge nodes, 200 IoT clients) show that FL-MSF achieves 96.5% detection accuracy, a 0.95 F1-score, and low latency (31 ms), outperforming baseline long short-term memory networks, graph neural networks, and FedAvg models.Gradient leakage risk was reduced to <0.5%, and scalability tests presented accuracy gains as sensor density increased.Performance under extreme scaling and management of long-term drifts in material attenuation parameters must be addressed to enhance the availability of the developed system.