Cybersecurity is a fast-changing field where data patterns are always shifting. Attackers develop new ways to break into digital systems and networks. Due to this ongoing threat, Intrusion Detection Systems (IDS) are vital for protecting modern cyber infrastructure. Recently, Machine Learning (ML) and Deep Learning (DL) based intrusion detection systems have shown notable improvements in spotting malicious activities and identifying cyber-attacks more accurately. Deep neural networks can learn complex patterns from large datasets, which leads to better detection compared to traditional methods. However, as these models become more accurate and complex, they also become harder to understand, making it challenging to trust and use them in real-world situations. Many deep learning models operate like “black boxes”; the reasoning behind their predictions is often unclear.

 

This research paper suggests an Explainable Artificial Intelligence (XAI) framework for Network Intrusion Detection Systems. The goal is to improve transparency and trust in machine learning-based security solutions. The framework combines deep neural networks with explainability techniques to give useful insights into the decision-making process of the model throughout different stages of the machine learning pipeline. The study uses the NSL-KDD dataset to assess the performance of this approach. Various XAI techniques, such as SHAP, LIME, Contrastive Explanations Method (CEM), ProtoDash, and Boolean Decision Rules via Column Generation (BRCG), are used to create explanations for the predictions made by the IDS model. These methods help identify which features have the most impact on detecting cyber-attacks and evaluate their influence on the final prediction. The results show that combining deep learning with explainable AI can improve both detection accuracy and model transparency in cybersecurity applications.

Keywords: classification, intrusion detection system, cybersecurity, explainability, SHAP scores, explainable AI, Deep neural network, SHAP, LIME, AIX 360, BRCG, CEM, ProtoDash, local explanations, global explanations, rules