Optimizing Intrusion Detection in Wireless Sensor Networks via the Improved Chameleon Swarm Algorithm for Feature Selection

In this paper, the improved chameleon swarm algorithm (ICSA) enhances the exploration–exploitation balance while optimizing feature subset selection. The integration of Lévy flight-based exploration refines ICSA's search strategy, complemented by rotation-type refinement and adaptive parameter-setting mechanisms. These modifications ensure that exploration aligns effectively with the feature selection process, leading to a more adaptive and efficient approach. To evaluate ICSA's effectiveness, it is tested on the NSL-KDD benchmark, a well-established dataset in intrusion detection systems. Performance is assessed based on key metrics, including accuracy, detection rate, false alarm rate, execution time, and the number of selected features. Comparative analysis against six advanced classifiers demonstrates that ICSA achieves superior results with minimal computational overhead. The algorithm attains the highest accuracy (97.91%) and detection rate (98.75%), the fastest execution time, and the lowest false alarm rate (0.0021), eliminating the need for excessive feature selection. These results confirm that modifying feature selection mechanisms within ICSA significantly enhances computational efficiency and detection performance, as validated through rigorous experimental testing at the classifier level.