Federated learning and edge computing-based collaborative detection system for IoT anomaly behavior

Abstract

This research proposes a collaborative detection system for IoT device anomalous behavior, integrating federated learning with edge computing. By integrating distributed machine learning with localized computational resources, the proposed framework enables collaborative detection characterized by minimal latency and robust data privacy. The system was experimentally validated using a real-world IoT dataset that encompassed 12 device types and over 2000 nodes, with data collected over a 30-day period. The system aggregates local model parameters from edge nodes within a federated learning framework, enabling model sharing while preserving data privacy, and achieves a detection accuracy of 96.3%, which is 21.5% higher than traditional centralized approaches. The edge computing module conducts local data preprocessing and feature extraction, reducing per-device detection latency to below 50 ms—a 68% improvement over cloud-based solutions. Energy consumption analysis indicates that the overall system energy usage is reduced by 32% compared to centralized methods, supporting extended device operation for up to 72 h. Through a multi-node collaborative training mechanism, the system maintains a recall rate above 92% even at a scale of 500 nodes, and the response time for identifying novel attack behaviors is shortened to within 15 min. This study demonstrates the efficacy of the synergy between federated learning and edge computing in IoT security contexts, providing a scalable and privacy-centric solution for anomaly detection across large-scale distributed devices.