Federated Learning-Enabled Intrusion Detection System for Resource-Constrained IoT Devices in Adversarial Environments

Abstract

The increasing deployment of Internet of Things (IoT) devices in security-critical and resource-constrained environments has amplified the demand for efficient and privacy-preserving Intrusion Detection Systems (IDSs). Traditional centralized IDSs fail to meet the real-time, lightweight, and privacy-aware requirements of modern IoT networks. This paper proposes a Federated Learning (FL)-enabled IDS architecture specifically designed for resource-constrained IoT devices facing adversarial threats such as Denial of Service (DoS), Man-in-theMiddle (MitM), spoofing, and malware injection or data poisoning. The proposed system employs decentralized training across IoT nodes while preserving local data privacy. Our model combines lightweight deep learning classifiers and robust aggregation strategies to ensure accuracy and efficiency. Experimental evaluations on benchmark datasets demonstrate high detection accuracy, reduced communication overhead, and strong resilience against evolving attack vectors, highlighting the viability of our FL-IDS in real-world IoT deployments.