An Enhanced Horse Herding Optimizer-Driven Feature Selection Approach Integrated with CNN for Network Intrusion Detection Systems

Intrusion Detection Systems (IDS) are essential for protecting networks from various types of malicious activity. Network Intrusion Detection Systems (NIDS) examine and evaluate the network traffic to detect unusual activity, possible threats, and attempts at illegal access to the network. However, due to the high-dimensional feature spaces, redundant information, and low detection accuracy frequently hinders their performance. Therefore, feature selection and dimensionality reduction are crucial parameters for improving the detection capabilities and efficiency. To overcome these challenges, this proposed work employs a Convolutional Neural Network (CNN) for intrusion detection, along with an Enhanced Horse Herding Optimizer (EHHO) for feature selection integrated with Kernel Principal Component Analysis (KPCA). Classification accuracy and feature subset size are two competing requirements that are balanced by the weighted-sum fitness function used by the EHHO, which enables the model to attain good detection performance at a low computational cost. Extensive experiments were carried out by using benchmark IDS datasets. The proposed EHHO-KPCA-CNN model outperformed other baseline techniques in terms of overall classification accuracy, false alarm rate of, and a detection rate. These findings demonstrate that the proposed model is scalable and provides efficient solution when it is compared with contemporary IDS applications.