A Comparison of Feature Selection and Feature Extraction in Network Intrusion Detection Systems

Abstract

Network intrusion detection systems (NIDS) are crucial components in securing computer networks. As network traffic data typically consists of numerous features, dimensionality reduction is often necessary to enhance detection efficiency. This paper presents a comprehensive comparison between feature selection and feature extraction techniques in network intrusion detection systems.

We evaluate several feature selection methods (including Information Gain, Chi-Square, Correlation-based Feature Selection) against feature extraction approaches (such as Principal Component Analysis, Linear Discriminant Analysis, and Autoencoder-based feature extraction) on benchmark network security datasets. Our experimental results demonstrate the effectiveness of both approaches in improving detection accuracy while reducing computational overhead.

The findings indicate that feature selection techniques generally maintain better interpretability of the features, while feature extraction methods achieve slightly higher performance in classification accuracy at the cost of reduced explainability. This trade-off analysis provides valuable insights for security practitioners when designing efficient and effective intrusion detection systems.

Keywords

Network Security, Intrusion Detection, Feature Selection, Feature Extraction, Machine Learning

Recommended citation: Vuong Tuan-Cuong, Tran Hung, Trang Mai Xuan, Ngo Vu-Duc, Luong Thien Van. (2022). "A Comparison of Feature Selection and Feature Extraction in Network Intrusion Detection Systems." 2022 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC), pp. 1798-1804.
Download Paper