An Integrated Knowledge Graph and Graph Neural Network Framework for Scalable Cyber Threat Attribution and Proactive Defense
Vinod B Maniyat, B. R. Arun Kumar · 2025
This paper presents ETRGNN-ZT, a scalable and automated cybersecurity framework that integrates Neo4jbased knowledge graphs, Graph Neural Networks (GNNs) using the Deep Graph Library (DGL), and Zero-Trust (ZT) policy-driven mitigation. The proposed framework addresses key limitations of static security models by introducing a realtime pipeline for threat detection, risk prioritization, and proactive mitigation. Structured data from the MITRE ATT&CK framework and unstructured Open Source Intelligence (OSINT) are fused to construct a dynamic knowledge graph representing attack tactics, vulnerabilities, and indicators of compromise (IOCs). The GNN module detects attack paths and generates threat scores, which drive automated ZT policy updates. Experimental evaluation demonstrates a detection accuracy of 96.7 %, a mitigation success rate of 98.4 %, and sub- 0.1 second inference latency, while maintaining linear scalability across millions of graph nodes. The ETRGNN-ZT framework offers an intelligent, resilient, and adaptive approach to modern cyber defense.