AI-Driven Threat Intelligence using Graph Neural Networks for Advanced Cybersecurity Defense
Ramana Rajendran, Venkata Pavan Kumar Reddy Chintham, R. Beaulah Jeyavathana, Vamsi Krishna Chidipothu, Venu Karunanithi, B. Jegajothi · 2025
The continuous growth of cyber threats needs sophisticated and adaptive defensive systems capable of proactively detecting malicious activity across complicated network architectures. This study presents TIGNET, a unique AI-driven threat detection architecture that combines sophisticated Graph Neural Networks (GNNs) with a bespoke feature improvement module called CAFE (Context-Aware Feature Extractor). TIGNET, which is designed to properly capture both structural and temporal relationships in network data, provides a multi-level reasoning framework for exact threat assessment. We test the model using the CSE-CIC-IDS2018 dataset, which comprises a wide range of attack scenarios spanning several days of simulated network activity. TIGNET outperforms baseline models such as SVM, Random Forest, LSTM, GCN, and GAT, with a superior accuracy of 98.12%, precision of 97.02%, recall of 97.89%, and an F1-score of 97.45%, thanks to thorough testing. The suggested model also has a low false positive rate (1.21%) and high detection rates for several attack types, including 99.45% for DoS Hulk and 98.91% for PortScan. The suggested architecture has a high potential for real-world application owing to its scalability, interpretability, and resilience. This study adds a realistic and new approach to the area of AI-powered cybersecurity by addressing both detection accuracy and dependability in dynamic threat environments.