Forecasting and Analysing Cyber Threats with Graph Neural Networks and Gradient Based Explanation for Feature Impacts

Cybersecurity is increasingly critical in today’s digital landscape, with evolving threats such as advanced persistent threats (APTs), ransomware, and phishing attacks posing significant risks to organizations and individuals alike. Traditional methods of malware detection have become insufficient as these threats grow in sophistication, highlighting the need for innovative approaches in threat management. The work addresses the gap in predictive and interpretative threat detection by focusing on AI-driven threat prediction. Graph Convolutional Network , one of the advanced Graph Neural Network (GNN) models are developed. The methodology includes building a website for file uploads, generating hash codes using Local Sensitive Hashing (LSH), training GNNs for threat detection. The gradient-based explanation method is involved from Explainable AI (XAI) that helps in interpreting the impact of individual features on a model’s predictions. This approach contributes to the field by enhancing proactive threat management and providing transparent, understandable explanations of detection results, ultimately advancing cybersecurity defences.