A Smart and Secured Anomaly Detection Scheme for Internet of Things (IoT) Powered Devices Using Federated Learning Technology

In the rapidly expanding realm of Internet of Things (IoT), ensuring the security of connected devices is paramount. Smart homes, which rely on an array of IoT devices like cameras, motion sensors, and door locks, are particularly vulnerable to anomalies that may indicate security breaches. This paper proposes a novel anomaly detection scheme leveraging Federated Learning technology to enhance the security of IoT -powered devices while maintaining data privacy. Unlike traditional models, which rely on centralized data collection, Federated Learning enables local anomaly detection models to be trained on each device without transferring raw data. The proposed model integrates AutoEncoder for dimensionality reduction and LSTM for detecting temporal anomalies in IoT environments. Evaluated against nine existing models, the proposed solution achieved an accuracy of 97.19%, significantly outperforming state-of-the-art models such as XGBoost (91.78%) and Deep Neural Networks (90.55%). Moreover, the model reduced communication overhead by 35% and showed a 180ms latency, making it highly efficient for real-time anomaly detection. This research demonstrates that Federated Learning can offer robust, scalable, and privacy-preserving solutions for anomaly detection in smart homes.