A METHOD FOR DATA DIMENSIONALITY REDUCTION IN NETWORK TRAFFIC CLASSIFICATION FOR INTRUSION DETECTION SYSTEMS
Abstract
The paper addresses the problem of high data dimensionality in network traffic classification, which significantly affects the accuracy, computational complexity, and real-time applicability of intrusion detection systems. Modern network traffic datasets contain a large number of features, many of which are redundant or correlated, leading to increased training time, higher risk of overfitting, and reduced generalization ability of machine learning models. Therefore, the development of dimensionality reduction methods that preserve informative features while decreasing computational cost is an important task in the design of efficient intrusion detection systems.The aim of the study is to develop a method for data dimensionality reduction in network traffic classification for intrusion detection systems that provides a balanced trade-off between classification quality and computational complexity. The proposed approach is based on the sequential application of mutual information for filter-based feature selection, principal component analysis for removing correlations and compacting the feature space, and a genetic algorithm for optimizing the subset of principal components.Experimental evaluation was performed using the UNSW-NB15 and NSL-KDD datasets with Random Forest as the baseline classifier and a unified preprocessing pipeline including categorical encoding and feature normalization. The results demonstrate that mutual information allows significant reduction of the feature set without loss of classification performance, while the additional use of principal component analysis and genetic optimization enables substantial compression of the feature space with acceptable F1-score and ROC-AUC values. The obtained results confirm that the proposed method can be effectively applied in resource-constrained environments and real-time network traffic analysis systems.The proposed approach can be used in the development of modern intrusion detection systems, network monitoring tools, and intelligent cybersecurity applications