Feature Selection for Anomaly-Based Intrusion Detection System Using Information Gain and Mutual Correlation

To avoid high computational costs inidentifying intrusions by IDSs, the size of adataset needs to be reduced. Feature selection isconsidered a problem of global combinatorialoptimization in machine learning, which reducesthe number of features, removes irrelevant, noisyand redundant data, and results in acceptableclassification accuracy. This paper proposes acombine filter method by using IG (informationgain) and Mutual Correlation for featureselection in NSL-KDD dataset. IG was used toselect important feature subsets from all featuresin the NSL-KDD dataset. The resulted featuresset are combined with Mutual correlation to getthe optimal reduced features set. Tests are doneon NSL-KDD dataset which is improved versionof KDD-99 dataset. The results show that thenumber of selected features is reduced from 41 to14 and correlated 10 features. The proposedmethod not only reduces the number of the inputfeatures and memory and CPU time but alsoincreases the classification accuracy