A Chaos-Enhanced Hybrid Metaheuristic Feature Selection for Intrusion Detection Using Artificial Neural Networks

Intrusion Detection Systems (IDSs) play a vital role in protecting modern networks against increasingly sophisticated cyberattacks. However, the high dimensionality and redundancy of network traffic features often degrade detection accuracy and increase computational overhead. To address these challenges, this paper proposes a novel chaos-enhanced hybrid metaheuristic feature selection algorithm combined with an Artificial Neural Network (ANN) classifier for efficient and accurate intrusion detection (OBL-CHGWO–WOA). The proposed approach integrates Opposition-Based Learning (OBL) for population initialization, a hybrid Gray Wolf Optimization–Whale Optimization Algorithm (GWO–WOA) to balance exploration and exploitation, and a chaos-based local search mechanism to enhance convergence and avoid local optima. Feature selection is guided by a multi-objective fitness function that simultaneously minimizes classification error and the number of selected features. The optimized feature subset is then used to train an ANN for intrusion classification. Extensive experiments are conducted on four benchmark datasets, NSL-KDD, KDD CUP99, CICIDS2017, and UNSW-NB15. Experimental results demonstrate that the proposed method consistently outperforms state-of-the-art metaheuristic algorithms, including GA, PSO, GOA, TLBO, and SSA, in terms of accuracy, sensitivity, specificity, F1-score, and Mean Squared Error (MSE), while selecting fewer features with higher stability. Moreover, convergence, complexity, and scalability analyses confirm that the proposed model achieves superior detection performance without increasing computational or memory costs, making it highly suitable for large-scale and real-world intrusion detection applications.