Abstract

The growing availability of diverse Internet of Things (IoT) devices has similarly established real-time intrusion detection as a critical topic for many IoT system security researchers. Traditional data intrusion detection systems often fail to effectively handle the dynamic and decentralized nature of these IoT environments. This paper proposes a new Federated Deep Graph Neural Network (DGNN) algorithm to support real-time intrusion detection in heterogeneous IoT systems. The proposed algorithm attempts to learn the complex relationships between devices in an IoT environment using graph-based deep learning while maintaining privacy through a federated learning environment. Striking a balance between federated learning and deep graph neural networks purposefully supports distributed learning while also intending to preserve privacy by not sharing raw data. The performance of the algorithm is evaluated on several real-world (IoT) datasets, which display markedly improved detection accuracy, precision, and recall performance compared to traditional data models. The framework is scalable, robust, and provides the opportunity for real-time protection against known and unknown intrusions when deployed in a large-scale IoT network environment.