Research On Hybrid Feature Selection Algorithm For Network Intrusion Detection

In recent years,network data becomes more and more complex,especially when the amount of data available for analysis is limited,the higher and higher feature dimension directly leads to the decrease of time efficiency and detection accuracy of intrusion detection.As a common dimension reduction method,feature selection has become an essential means to improve the performance of intrusion detection.At present,a single type of feature selection method can not meet the detection performance requirements.The hybrid feature selection method which combines multiple methods provides a new solution.However,hybrid feature selection still faces the challenge of improving performance and detection accuracy.Based on the above problems,three hybrid feature selection schemes are proposed in this paper.The key analysis content and innovation points can be summarized as below.(1)To solve the problem of low detection accuracy caused by redundancy of network data features,the concept of redundancy sensitive values is proposed to judge the redundancy of features.Based on the redundancy sensitive value and recursive feature addition algorithm RFA,a redundancy sensitive hybrid feature selection algorithm SRFA is proposed.The algorithm first selects the feature with the highest ranking coefficient and adds it to the feature subset.Then,in the process of finding the feature with the highest ranking coefficient,it calculates the redundancy sensitivity value of the current feature and adds the feature with the minimum redundancy sensitivity value to the feature subset.Compared with RFA algorithm,the subset of features selected by SRFA algorithm has better detection performance and accuracy improved by 9.3%.(2)Aiming at the problem of too long feature selection time caused by the high feature dimension of network features,this paper proposes a parallel segmented redundancy sensitive hybrid feature selection algorithm PSRFA based on the SRFA algorithm.This algorithm uses the idea of segmentation and parallel processing.First,the original feature set is divided into feature segments with the same number of features.Second,the SRFFA algorithm is executed in parallel for each feature segment,and finally the feature selection results on each segment are summarized.The experimental data confirm that PSRFA algorithm does well in performance,and the time efficiency of feature selection is improved by more than 86% compared with the non-parallel SRFA algorithm.(3)In order to solve the problem of reduced detection accuracy caused by unbalanced network data,this paper proposes a weighted sensitive hybrid feature selection algorithm WSRS based on cooperative redundancy.The algorithm uses the class distribution of data to improve the redundancy sensitive value,and combines the cooperative relationship of features with the ranking coefficient as the discriminant index of feature importance.At the same time,the useless features are deleted in the process of selecting features with high importance and low redundancy.The experimental results show that the detection rate of a few samples,the detection accuracy of the data and the efficiency of feature selection time are obviously improved compared with the contrast algorithm.