Machine learning-based intrusion detection: feature selection versus feature extraction

Abstract

With the rapid growth of networked systems, the importance of effective intrusion detection systems has become increasingly critical. Machine learning techniques have proven valuable in this domain, but the high-dimensional nature of network data presents significant challenges. This research article provides an in-depth analysis comparing feature selection and feature extraction methods within the context of machine learning-based intrusion detection systems.

We conduct extensive experiments using multiple benchmark datasets to evaluate the performance of various dimensionality reduction techniques. Our comprehensive analysis covers computational efficiency, detection accuracy, and the interpretability of resulting models. We further examine how these techniques perform under different network conditions and attack scenarios.

The results demonstrate that while both approaches offer significant improvements in computational efficiency and performance, they present different trade-offs. Feature selection methods maintain semantic meaning of the original features and provide better explainability, whereas feature extraction techniques often achieve higher detection accuracy with more compact representations. These findings provide important insights for cybersecurity practitioners and researchers designing robust intrusion detection systems for diverse network environments.

Keywords

Network Security, Intrusion Detection Systems, Machine Learning, Feature Selection, Feature Extraction, Dimensionality Reduction

Recommended citation: Ngo Vu-Duc, Vuong Tuan-Cuong, Van Luong Thien, Tran Hung. (2023). "Machine learning-based intrusion detection: feature selection versus feature extraction." Cluster Computing, pp. 1-15.
Download Paper