A Federated Learning Framework with Self-Attention and Deep Reinforcement Learning for IoT Intrusion Detection

Document Type

Article

Publication Date

4-26-2025

Abstract

In this research, we introduce a novel Network Intrusion Detection System (NIDS) called FedAtten-DRL, designed specifically for IoT networks. This system addresses the challenge of detecting attacks in these networks, particularly new or zero-day threats that traditional NIDS struggle to identify due to insufficient and imbalanced training data. Our framework uses Deep Reinforcement Learning (DRL) to build more adaptive models that enhance detection capabilities. To boost initial performance, we incorporate pre-trained layers from supervised learning and apply an additional attention mechanism on key features, improving the system’s ability to detect complex attacks. Furthermore, federated learning (FL) is employed to reduce communication overhead and protect data privacy by enabling collaborative training across devices. We evaluate the system on real-time datasets, specifically Edge-IIoTset, which simulates critical security challenges in edge-based industrial IoT environments. The results demonstrate that our framework supports continuous learning and improves detection performance, even against evolving threats in resource-constrained IoT domains.

Share

COinS