Fortifying network security: machine learning-powered intrusion detection systems and classifier performance analysis

Abstract

Intrusion detection systems (IDS) protect networks from threats; they actively monitor network activity to identify and prevent malicious actions. This study investigates the application of machine learning methods to strengthen IDS, explicitly emphasizing the comprehensive CICIDS 2017 dataset. The dataset was refined by implementing stringent preprocessing methods such as feature normalization, class imbalance management, feature reduction, and feature selection to ensure its quality and lay the foundation for developing robust models. The performance evaluation of three classifiers-support vector machine (SVM), extreme gradient boosting (XGBoost), and naive Bayes was highly impressive. Vital accuracy, precision, recall, and F1-score values of 0.984389, 0.984479, 0.984375, and 0.984304, respectively, were achieved by SVM. Notably, XGBoost demonstrated exceptional performance across all metrics, attaining flawless scores of 1.0. naive Bayes demonstrated noteworthy accuracy, precision, recall, and F1-score performance, which were recorded as 0.877392, 0.907171, 0.877007, and 0.876986, respectively. The results of this study emphasize the critical importance of preparation methods in improving the effectiveness of IDS via machine learning. This further demonstrates the potential of particular classifiers to detect and prevent network intrusions efficiently, thereby substantially contributing to cybersecurity measures.