A Hybrid Approach for Feature Selection Analysis on the Intrusion Detection System Using Naive Bayes and Improved BAT Algorithm

In the last two decades, global internet usage has expanded to over 4 billion users, leading to a concurrent exponential rise in malicious network activities. To safeguard networks, Anomaly Detection Systems (ADS) are deployed; however, their efficiency is frequently hampered by high-dimensional data containing duplicate or irrelevant features. This paper proposes a novel feature selection method, the Naive Bayes Improved BAT Algorithm (NB-IBA), which utilizes entropy-based concepts to identify optimal feature subsets. The proposed method is evaluated using the CICIDS2017 dataset across multiple classifiers, including J48, Random Forest, Random Tree, and Bayesian Networks. Experimental results indicate that feature reduction significantly enhances performance, with the Random Forest classifier achieving superior accuracy across various attack vectors 

Keywords: Intrusion Detection System (IDS), Feature Selection, Swarm Intelligence, BAT Algorithm, Naive Bayes, Machine Learning.