An Advanced Ensemble Framework Employing Grey Wolf Optimization and Feature Selection Techniques for Enhanced Intrusion Detection on Unbalanced NSL-KDD Data

Abstract

Intrusion Detection Systems (IDSs) usually face severe issues with imbalanced datasets and the limited ability of a single classifier to generalize well. This research proposes a sophisticated ensemble method combining cutting-edge ensemble learning techniques with Grey Wolf Optimization (GWO), a recent metaheuristic optimization algorithm, and appropriate feature selection methods to significantly improve the accuracy of IDS. The framework is validated via the NSL-KDD dataset, proving that the stacking and voting ensemble methods proposed outperform stand-alone classifiers by a great margin. The stacking model optimized with Decision Trees and K-Nearest Neighbors classifiers as constituent models attains a superb F1-score of 98.9%. Ensemble methods optimized via GWO are highlighted in this work as effective, providing new insights and significant improvements for real-world utilization in intrusion detection systems.