AN EXPLAINABLE AI FRAMEWORK FOR NETWORK INTRUSION DETECTION USING MULTI-LAYER PERCEPTRONS

Abstract

Intrusion detection systems (IDS) are critical for protecting network infrastructure from nefarious activity. The growing use of machine learning models, especially Multi-Layer Perceptrons (MLPs) in intrusion detection systems (IDS), highlights the essential need of transparency and interpretability. This work seeks to improve the interpretability of MLP-based IDS models by investigating the incorporation of two Explainable Artificial Intelligence (XAI) methodologies: LIME (Local Interpretable Model-agnostic Explanations) and SHAP (SHapley Additive exPlanations). We evaluate LIME and SHAP using conventional IDS datasets to ascertain their effectiveness in clarifying model predictions in an understandable way for individuals. To improve the reliability and accountability of AI-driven security systems, our findings suggest that both techniques provide significant insights into model decision-making. The results illustrate the practical use of explainable artificial intelligence (XAI) approaches to improve the visibility and interpretability of black-box neural network models for cybersecurity experts.