An Enhanced Hybrid Genetic-JAYA Algorithm for Feature Selection and SVM Parameter Optimization in Intrusion Detection Systems: Evaluation on the CICIDS Dataset
Mohanad Ghazi Yaseen, Ahmed Hussein Ali, Mohammad Aljanabi · Shifra. · 2025
The changing scenario in the cybersecurity field requires intelligent and adaptive means of detecting sophisticated intrusions. This paper presents an improved hybrid Genetic-JAYA algorithm for simultaneous feature selection and hyperparameter tuning of an SVM-based IDS. Leveraging the CICIDS2017 dataset, which is well-known for its comprehensive representation of modern attack types, the model interprets the classification as a binary problem, distinguishing normal traffic from malicious activity. The presented hybridization employs the global exploration ability of the Genetic Algorithm (GA) with the parameter-less, convergence-oriented behavior of the JAYA algorithm. To select the subset of features, a custom fitness function is adopted that balances classification accuracy and feature compactness, while the hyperparameters of the SVM using a radial basis function (RBF) kernel are optimized concurrently for improved detection. All results were obtained using stratified 5-fold cross-validation, with metrics averaged over 10 independent runs. The hybrid GA-JAYA-SVM model outperformed GA-SVM and JAYA-SVM, achieving an average accuracy of 98.73% (±0.45), precision of 98.63%, recall of 98.79%, and a reduced false positive rate of 1.15%. The robustness and generalization capability of the model are further supported by AUC and F1-score measurements. Statistical evidence, via the Wilcoxon signed-rank test, validates the observed performance gains (p < 0.05). These results illustrate the capability of the proposed model to enhance IDS performance with minimal computational overhead an essential requirement for real-time deployment. Overall, the findings suggest that hybrid metaheuristic approaches, such as the Genetic-JAYA combination, offer strong potential for improving IDS effectiveness and form a promising foundation for future research into adaptive, intelligent, and layered cybersecurity defenses.