The Dataset Features Selection for Detecting and Classifying Network Attacks

This article focuses on the current problem of reducing the dimensions of datasets. Solving this problem will improve the speed and quality of response of intrusion detection systems (IDS) to network attacks. In order to reduce the dimensionality of the data, selection and feature generation methods are used using the example of CSE-CIC-IDS2018 dataset. The artificial neural network (ANN) multilayer perceptron is used as the intrusion detection system classifier. The main stages of data preprocessing in machine learning models as applied to the CSE-CIC-IDS2018 dataset are presented. Two methods of dataset feature selection are proposed: based on informativeness (Gini index) and based on correlation (Pearson coefficient). A comparison of the datasets obtained by the selection of informative features of the CSE-CIC-IDS2018 dataset with the original dataset is performed