An Intrusion Detection Feature Selection Method Based on Improved Mutual Information

Abstract

Feature selection methods for classification are crucial for intrusion detection techniques using machine learning. High-dimensional features in intrusion detection data affect computational complexity, consume more used resources and more time for data analysis, and the irrelevant and redundant features among them often hinder the performance of classifiers and mislead the classification task. Therefore, it is challenging to select more relevant features from intrusion detection data containing many such features. In this paper, we propose an efficient feature selection algorithm that first considers the correlation between features and the redundancy of pairs of features with respect to class labels based on an improved Pearson correlation coefficient, and later improves the evaluation function based on conditional mutual information to obtain a final subset of features with the goal of improving the classification rate and accuracy. The proposed feature selection method based on improved conditional mutual information is compared with three existing feature selection methods on the frequently studied public benchmark intrusion detection dataset NSL-KDD. The experimental results show that the features selected by the proposed method in this paper lead to a significant reduction in execution time while resulting in higher classification accuracy.