Correlation-Based Feature Selection for Intrusion Detection Design

Abstract

In a large amount of monitoring network traffic data, not every feature of the data is relevant to the intrusion detection task. In this paper, we aim to reduce the dimensionality of the original feature space by removing irrelevant and redundant features. A correlation-based feature selection algorithm is proposed for selecting a subset of most informative features. Six data sets retrieved from UCI databases and an intrusion detection benchmark data set, DARPA KDD99, are used to train and to test C4.5 and naive bayes machine learning algorithms. We compare our proposed approach with two correlation-based feature selection algorithms, CFS and FCBF and the results indicate that our approach achieves the highest averaged accuracies. Our feature selection algorithm could effectively reduce the size of data set.