Volume 18, Number 1
Advanced Graph Neural Network Mechanisms for IoT Network Intrusion Detection: A Comprehensive Study of Prototype-Based, Contrastive, and Structure Learning Approaches
Authors
Mehmet Baris Yaman, Foundational Technology, Turkey
Abstract
The proliferation of Internet of Things (IoT) devices has created unprecedented security challenges, with traditional intrusion detection systems struggling to achieve the accuracy needed for operational deployment. This comprehensive study investigates three advanced Graph Neural Network (GNN) mechanisms for network intrusion detection, addressing the fundamental limitations of standard edge-level classification approaches. We present Prototype-GNN, which leverages distance-based lassification with learnable prototype embeddings; Contrastive-GNN, which optimizes embedding geometry through supervised contrastive learning; and GSL-GNN, which adaptively learns optimal graph structure from node features. Through extensive experimentation on the TON-IoT dataset containing 1 million network connections, we demonstrate that these mechanisms achieve 94.24%, 94.71%, and 96.66% accuracy respectively, representing substantial improvements of +2.37, +2.84, and +4.79 percentage points over the baseline EdgeLevelGCN (91.87%). Our best-performing GSL-GNN architecture achieves 99.70% ROC-AUC with an exceptionally low 1.5% false positive rate, addressing the critical challenge of alert fatigue in security operations. This journal article extends our preliminary conference study [1] by providing comprehensive ablation studies, additional architectural variants, and deeper theoretical analysis.
Keywords
Graph Neural Networks, Network Intrusion Detection, IoT Security, Prototype Learning, Contrastive Learning, Graph Structure Learning, Deep Learning