An ensemble framework with improved hybrid breeding optimization-based feature selection for intrusion detection

Intrusion detection is a proactive means to detect network attacks and has been a hot point in network security. To address the curse of dimensionality and improve the Intrusion Detection System (IDS) performance, Hybrid Breeding Optimization (HBO), a novel metaheuristics algorithm inspired by the Chinese three-line hybrid rice breeding process, was implemented in IDS and has achieved good performance. However, it lacks adaptability and tends to get stuck in the local optimum during instantiation. Therefore, this study proposes a novel ensemble framework with improved HBO-based feature selection (FS) for intrusion detection. More specifically, the essential HBO is first modified by levy flight and elite opposition-based learning strategies (LE-HBO) to enhance its ability to seek the optimum. In addition, to make LE-HBO better applied to FS for intrusion detection, a Cooperative Co-evolution Improved HBO (CCIHBO) is proposed. It ranks and groups the features in the data samples, assigns subpopulations of LE-HBO of the appropriate size to each feature space, and finds the optimal feature subset through collaborative cooperation among the subpopulations. Finally, the proposed approach is implemented in benchmark datasets CEC2021, UCI, and security datasets NSL-KDD, WUSTL-IIOT and HAI datasets, in which KNN, SVM, and XBGoost are employed as classifiers for intrusion detection. Experimental results demonstrate that the proposed framework can effectively improve the accuracy of intrusion detection and outperform state-of-the-art methods in relevant evaluation indicators.