An Improved Feature Selection Approach Using Black Widow Optimization

Abstract

Intrusion Detection Systems (IDS) play a critical role in safeguarding modern computer networks against increasingly sophisticated cyber-attacks. However, the high dimensionality and redundancy of network traffic features often degrade detection performance and increase computational overhead. To address this challenge, this paper proposes a Black Widow Optimization (BWO)-based wrapper feature selection framework for intrusion detection using the NSL-KDD dataset. The proposed approach aims to identify a compact and discriminative feature subset that maximizes classification performance while minimizing feature dimensionality. BWO exploits unique biological mechanisms, including sexual and sibling cannibalism, to effectively balance exploration and exploitation during the search process. Experimental evaluation demonstrates that the proposed method reduces the original 41 features to 15, achieving a feature reduction of 63.41% while improving classification accuracy to 96.01%. In addition, the false alarm rate is reduced to 2.89%, indicating enhanced detection reliability. Confusion matrix analysis confirms high detection capability for both normal and attack traffic, and statistical significance testing (p-value = 0.0038) validates the robustness of the observed performance improvements. The results indicate that Black Widow Optimization is an effective and competitive feature selection strategy for intrusion detection systems, offering improved accuracy, reduced false alarms, and lower computational complexity.