Enhanced teaching–learning optimization algorithm for feature selection in intrusion detection systems

The growing complexity and volume of network traffic have further emphasized the necessity for reliable and efficient intrusion detection systems. The presence of redundant and noisy data in the feature space complicates intrusion detection and increases computational complexity. Consequently, feature selection has emerged as a vital component in the design of intrusion detection systems, and wrapper-based techniques have been shown to be effective for improving their intrusion-detection capability when combined with the optimization power of metaheuristic techniques. Although the teaching-learning-based optimization algorithm has proven effective for feature selection, its previous variants are plagued by premature convergence, origin bias, and limited population diversity when optimizing complex intrusion detection problems. To overcome the aforementioned issues associated with existing variants of the teaching-learning-based optimization algorithm, the present study proposes a wrapper-based IDS that employs the golden-sine-guided multi-population teaching-learning optimization for feature selection. The proposed technique leverages golden-sine-guided optimization and multi-population learning to overcome the aforementioned issues associated with the teaching-learning-based optimization. A decision tree classifier is included in the proposed method to determine the feature subset. The results from experiments on the KDDCUP99, UNSW-NB15, and NSL-KDD datasets show that the suggested technique yields higher accuracy, reduces false positives, and selects the smallest feature subset compared to existing optimization techniques.