A Theoretical Analysis of Feature Selection Methods in Machine Learning-Based Intrusion Detection Systems

Abstract

With the increasing complexity of the network environment, more and more machine learning-based Intrusion Detection Systems (IDS) are now being applied to identify unusual behaviour in network traffic. Many Dimensions and excess Redundancy reduce Detection Accuracy and Computation Efficiency. Therefore, to enhance the generalisation power of the model, feature selection can be employed. This paper offers theoretical research on feature selection for machine learning-based intrusion detection systems. Systematically review the traditional machine learning algorithms of Support Vector Machines (SVM), Decision Trees, Random Forests, etc., and explore their applications in intrusion detection. The three categories of feature selection are shown in more detail below: filter methods, wrapper methods and embedded methods. The change in detection accuracy and false-positive rate, as well as computational efficiency and model stability, are all related to feature selection. The study shows that feature selection plays a critical role in improving intrusion detection performance by reducing dimensionality, eliminating redundant features, and enhancing model generalization. The aim of this paper is to provide comprehensive theoretical support for the investigation of cooperation between feature selection methods and machine learning models in intrusion detection systems, and to offer a new reference for future research on high-efficiency and intelligent cybersecurity solutions.