Lightweight & Robust Federated Learning for IOT Anomaly Detection

Abstract

The increasing growth of Internet of Things (IoT) devices brought up substantial security concerns due to distributed synthetic data simulation, resource constraints, privacy concerns, and increasingly sophisticated cyberattacks. To overcome these issues, a horizontal federated learning approach is used to present a lightweight and secure federated learning (FL) framework for IoT anomaly detection that ensures that multiple IoT devices jointly learn a universal model as maintaining local data privacy. A lightweight autoencoder architecture is designed for low-power IoT devices, which facilitating effcient local anomaly detection based on reconstruction error with low computational and memory overhead. To enhance robustness for more realistic non-IID and heterogeneous data distributions, the federated learning method implements robust aggregation technique, such as median aggregation, to effectively reduce negative effects of malicious or unreliable client updates. By using benchmark datasets - BoT-IoT and TON-IoT, which simulate the realistic IoT environments with adversarial and malicious clients, the presented framework is thoroughly evaluated. Experimental results shows that the proposed federated learning approach outdoes isolated local training, and robust aggregation techniques effectively reduce the negative effects of poisoned or malicious updates, which compared to the standard FedAvg algorithm, ensuring stable convergence and high detection accuracy. Finally, this framework successfully closes the robustness, effciency, and privacy gaps, by delivering effcient and secure solution for detecting anomalies in real-world IoT environments. Additionally, such methods is highly scalable as more IoT clients gets engaged in large-scale IoT networks. The communication-effcient design further reduces overhead during federated training, which is critical for bandwidth-constrained IoT networks, It spotlight the viability of the proposed framework for secure, feasible, and resilient deployment in real-world distributed IoT environments.