The Optimization Research Of Feature Selection Based On Mutual Information In Intrusion Detection

With the rapid development of computer and Internet technology,the intrusion behavior that threatens and destroys computer and network security emerge in endlessly.Intrusion detection system is a tool for detecting abnormal activities in computer and network,and is one of the effective ways to achieve higher security.In the face of the current surge in network traffic,the efficiency and precision of the intrusion detection system will decrease significantly when dealing with these high dimensional large data sets.Therefore,in order to solve this problem,this paper applies the feature selection method to intrusion detection.Feature selection is one of the important techniques of machine learning,data preprocessing by means of feature selection,excluding some noise data redundant and irrelevant data,the high dimensional data into low dimensional data.Then the streamlined data sets used for training the classifier,finally to improve the classification the performance and efficiency of the intrusion detection.The classic feature selection algorithm MIFS uses mutual information(MI)as a metric for evaluating feature subsets.The optimal feature subset is selected by calculating the maximum value of the mutual information between the feature and the target class and eliminating the redundancy between the selected feature and the candidate feature.Because the MIFS algorithm does not take into account the impact of the number of input features on the correlation between features,the modified mutual information feature selection algorithm MMIFS is optimized based on the MIFS algorithm.The MMIFS algorithm adopts the "minimum redundancy and maximum relevance" standard which is famous for evaluating feature subset.The purpose is to maximize mutual information between target categories and features,and minimize redundancy among features.But these two algorithms all need to use a indeterminate proportional coefficient to correct the redundancy between features,and choose a suitable parameter without a criterion.If the set value of the parameter is incorrect,the result of feature selection will be affected.Therefore,these two algorithms have some limitations.After deeply researching the above two kinds of mutual information feature selection algorithms,In this paper,an optimized mutual information feature selection algorithm OMIFS is proposed,and an intrusion detection system is established by combining the LSSVM classifier.we using MATLAB platform to compare the proposed algorithm OMIFS with the improved mutual information based feature selection algorithm MMIFS and the linear correlation selection algorithm LCFS in the NSL-KDD intrusion detection data set.The experiment mainly compares and analyzes the classification performance of different feature selection algorithms after data feature selection,and applies it to the classification performance of the same classifier LSSVM,and compares the performance of OMIFS+LSSVM based IDS and different classification algorithms applied to IDS.The simulation results show that,in the NSL-KDD data set,the OMIFS algorithm is better than the other two feature algorithms to improve the classification performance,and compared with IDS based on SVM algorithm and Clustering algorithm,IDS based on OMIFS+LSSVM has better intrusion detection performance.