Unlocking the Potential of Naïve Bayes for Network Intrusion Detection: A RandomForest-Driven Feature Selection Strategy
Rayane Moustapha ElRaba'a, Soha Rawas, Ali El‐Zaart · 2023
Network intrusion detection systems play an essential role for ensuring an organization’s security success. This study focuses on the most recent intrusion detection systems built using the efficient and straightforward Naïve Bayes machine learning technique. However, in high-dimensional, imbalanced data spaces, such Machine Learning-based systems often underperform. To address this, feature selection approach is implemented using SelectKBest with Statistical Tests, a RandomForest-based filter, and Permutation Importance on the CSE CIC IDS 2018 dataset. The workflow begins with data transformation using SelectKBest with a chi-square test then applies the suggested method. Subsequently, feature relevance with Permutation Importance is assessed, eliminating low-impact features. Experiments include evaluation analysis using multiple evaluation metrics on Naïve Bayes classifier, showing that the RandomForest-based feature selection method significantly boosts Naïve Bayes’ overall performance. This leads to enhanced existing anomaly detection algorithms, and improved utilization of resources, underscoring the research’s introduction of an innovative feature selection method with a groundbreaking impact on the effectiveness and precision of intrusion detection systems, strengthening overall Security.