Feature Selection with Improved Mountain Gazelle Optimizer Algorithm For Intrusion Detection Systems

Abstract

This master's thesis, titled "Feature Selection with Improved Mountain Gazelle Optimizer Algorithm for Intrusion Detection Systems," presents a comprehensive evaluation of the effectiveness of the Improved Mountain Gazelle Optimizer with Quasi-Oppositional Based Learning (MGO-QOBL) in optimizing feature selection for various classifiers. The study systematically compares MGO-QOBL with traditional optimization algorithms such as Ant Colony Optimization (ACO), Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and the standard Mountain Gazelle Optimizer (MGO) across multiple performance metrics, including accuracy, precision, F1 score, and runtime. The results indicate that MGO-QOBL consistently delivers superior or highly competitive performance across different classifiers. MGO-QOBL significantly enhances precision and accuracy for classifiers like the Decision Tree Classifier (DTC) and Random Forest Classifier (RFC) while maintaining robust performance for GaussianNB and SVM. Regarding F1 scores, MGO-QOBL demonstrates a balanced improvement, combining high precision and recall. Despite increasing computational costs, the trade-off with improved performance metrics justifies the increased runtime, particularly for GaussianNB and KNN classifiers. Statistical validation using the Wilcoxon signed-rank test further reinforces the reliability of these findings, showing significant improvements in many cases. These results underscore the efficacy of MGO-QOBL in feature selection for intrusion detection systems, making it a valuable optimization tool compared to traditional methods. The study highlights the potential of MGO-QOBL to advance the accuracy and reliability of intrusion detection systems, contributing to more secure and efficient cybersecurity infrastructures. Future research directions include evaluating the performance of MGO-QOBL across different datasets, exploring hybrid optimization approaches, and applying the algorithm to real-time intrusion detection systems and other domains such as bioinformatics and finance. This thesis demonstrates that MGO-QOBL is a powerful tool for enhancing feature selection processes, with significant implications for the broader machine learning and optimization field.