CYBERSECURITY THREAT MODELING USING GRAPH NEURAL NETWORKS

Abstract

Graph Neural Networks (GNNs) have emerged as a powerful approach for modeling complex relational data, which makes them well suited for cybersecurity threat modeling—where network relationships, entity interactions and propagation paths define attack surfaces and vulnerabilities. This paper investigates how GNN-based techniques can be applied to threat modeling in cybersecurity, covering architecture, feature extraction, evaluation metrics and deployment considerations. We present two illustrative charts: one showing detection accuracy vs model complexity, and another comparing incident response time vs graph model adoption. We review empirical investigations, benchmark results and casestudies, identify challenges in data collection, scalability and explainability, and propose a research roadmap for deploying GNNdriven threat models in enterprise settings. Our analysis shows that GNNbased models offer improvements in detection accuracy, generalisation to novel threats and ability to model multistep attack pathways—provided that graph construction and feature engineering are done carefully.