Leveraging Explainable AI for Pointer Usage-based Software Vulnerability Analysis

Deep learning has greatly enhanced software vulnerability analysis by uncovering complex patterns, enabling more accurate detection and analysis. However, these models often lack transparency, making their predictions difficult to interpret. To overcome this limitation, Explainable AI (XAI) methods have gained importance, as they provide deeper insights into model decisions when integrated with machine learning. This paper examines popular XAI techniques for features interpretation. The aim is to enhance the interpretability of software metrics related to vulnerabilities, especially those involving pointer usage. By applying SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) to standard software metrics, we rank features that influence vulnerability detection and provide insights into the factors contributing to software vulnerabilities. This approach enhances the transparency of detection results, making them more interpretable and actionable for developers and security analysts.