A Hybrid Deep Learning Framework for Enhanced Intrusion Detection in IoT Networks: Integrating Federated Learning and Attention Mechanisms

Abstract

The exponential growth in Internet of Things (IoT) devices has opened up an enormous and exposed attack surface, necessitating stronger intrusion detection systems (IDS). Conventional centralized IDS solutions are not scalable and do not ensure privacy for the distributed IoT paradigm. This paper introduces a new hybrid deep learning model combining federated learning and attention mechanisms to provide improved intrusion detection for IoT networks. The framework capitalizes on the decentralized aspect of federated learning to learn a global model cooperatively across IoT devices without exposing sensitive information. Attention mechanisms are built into the deep learning framework to attend to the most informative features for effective anomaly detection. We apply and test the designed framework on a benchmark IoT intrusion detection dataset, achieving substantial improvement in detection accuracy, minimizing communication overhead, and privacy enhancement over current state-of-the-art techniques. The results highlight the promise of this hybrid solution for designing robust and privacy-compliant security solutions for the fast-growing IoT domain.