Federated Learning For Anomaly Detection In Internet Of Things Environments

Abstract

This study develops a novel edge-based anomaly detection system for IoT environments utilizing federated learning (FL) with Random Forest algorithm within a federated averaging (FedAvg) framework, emphasizing privacy preservation and real-time detection. The methodology employed involves locally training models on data collected from diverse IoT devices, which only model updates are send to a central server to collaboratively improve global model accuracy without exposing raw data. The study employed the Edge-IIoTset dataset for experimental evaluation, measuring performance through accuracy, precision, recall, F1-score, and macro F1. The evaluation results shows that the federated model achieves 97% binary classification accuracy and exhibits robust multi-class detection, superior to traditional centralized models, while significantly reducing communication overhead and safeguarding data privacy. The evaluation across multiple edge devices confirms the system’s suitability for real-time, resource-constrained environments, addressing existing gaps in IoT anomaly detection by offering a scalable, secure, and effective decentralized approach. The study also validates that integrating RF algorithm with FL provides a promising solution for anomaly detection that balances high performance with privacy, outperforming existing centralized methods and advancing the deployment of secure IoT systems in cybersecurity applications.