An Optimized Intrusion Detection Model Using ML and Explainable AI
Nikunj Jain, Jawed Hawari, Priyanshu Jha, H. N. Vishwas, Manish Jain · 2024
In the maintenance of network security, Intrusion Detection Systems (IDS) are very crucial in identifying unauthorized access and malicious activities. This research assesses how well different machine learning algorithms enhance IDS functionality, highlighting model interpretability through explainable AI techniques. Eight machine learning algorithms were assessed using a dataset from Kaggle that reflects a military network environment with 41 features and 22,544 records. Local Interpretable Model-agnostic Explanations (LIME) was used to analyze the top five models for their decision-making procedures. In several metrics, Random Forest algorithm appeared as the best performing among them all. LIME application provided useful insights into the main features that impact on the predictions being made by the model. The outcomes demonstrate the effectiveness of Random Forest for IDS and indicate how vital explainable AI methods such as LIME are in ensuring that IDS deployment can achieve transparency and reliability concerning its predictions for this model’s purpose.