Research On Intrusion Feature Selection Methods Based On ReliefF-FCBF Combination

In the information age, a lot of new network data into the Internet every day, the abnormal behavior of the network space is more complex and changeable, the high dimension of data attributes, resulting in the detection efficiency of intrusion detection,the accuracy rate is low. In order to build a safe and sustainable network environment,to provide protection for the rapid development of the Internet, intrusion detection system to introduce new technology to achieve innovation without delay. Commonly used feature selection methods is introduced to intrusion detection system, a preliminary choice of network data, to the multidimensional data dimensionality reduction, to remove the irrelevant, weakly correlated and redundant features,improve classification efficiency, accurate rate, reduce the false alarm rate, improve the performance of intrusion detection system, realization of the intelligent system and meet the modern network space safety testing requirements.This reference to the characteristics of domestic and foreign selection algorithm in Intrusion Detection Application Research of typical four feature selection algorithm,through the advantages of complementary set two new, dimensionality reduction method for intrusion detection system of KDD cup 1999 data sets. In this paper, do the following work:(1) the comparison of four algorithms: ReliefF algorithm can not distinguish the selected feature set features between closely related sexual; fcbf algorithm can efficiently handling characteristics between redundancy; Re-ReliefF algorithm in efficiency, accuracy, false alarm rate is not good enough; among features in the correlation of the maximum correlation algorithm for minimum redundancy(mrmr)representation of the mutual information measure criterion, can better inter feature correlation.(2) improved by the combination of two algorithms: ReliefF algorithm and fcbf algorithm is combined with, put forward the two stage Re-FCBF algorithm,respectively, the original feature set the corresponding screening, distinguished by thecomposition data, the characteristics of the optimal feature subset; algorithm Re-ReliefF+ in Re-ReliefF algorithm based joined the idea of maximum correlation algorithm for minimum redundancy(mrmr). The feature correlation expressed as mutual information criterion is introduced into the, so as to better area divided into relationship between features and further remove the redundant features.(3) improved combined algorithm experimental analysis: the KDD cup 1999 as intrusion detection data, using support vector machine(SVM) to the training data and test data, corresponding optimal feature subset in data classification accuracy, false positive rate, false negative rate and training time and forecast. Experiments show that a valid filter feature selection algorithm is Re-FCBF algorithm and Re-ReliefF+, it can effectively remove irrelevant and weakly correlated and redundant features,improve the intrusion detection efficiency and accuracy, reduce the rate of false positives.