Improved Feature Selection Method using SBS-IG-Plus

Abstract

Any network can be represented by a set of attributes (features). There are several methods attempting to derive the best feature set at all. In our proposed method we exploited the anomaly detection rate and the false positive rate by using a Sequential Backward Search (SBS) method followed by Information Gain (IG) to get the best and valuable feature set properly. In SBS we used several classifiers to evaluate our results. These are Neural Network, Naïve Bayes, Decision Tree and Random Tree. Our method outperformed other approaches because of involving SBS method. We could reduce the huge data set to a more efficient and usable data set without any irrelevant features that defer the performance. Using these features we have enhanced the detection rate and reduced the alarm rate in the network. We have chosen the improved dataset from DARPA 98 Lincoln Lab evaluation Data set (NSL-KDD Set), because it is universally used for learning and testing in the IDS.