Research On Feature Selection Method For Network Intrusion Detection

With the rapid development of network technology,there are massive data in the network environment,and network security issues are particularly important.Insecure network environment may lead to many problems such as privacy leakage and resource theft,which brings many losses to people's work and life.Therefore,network intrusion detection is becoming a research hotspot.Network intrusion detection can find out whether there are any violations of security policy and signs of being attacked by analyzing network information.The feature selection of network intrusion data is a crucial part of network intrusion detection,which directly affects the detection effect.The data of network intrusion detection has the characteristics of large data volume and high dimensionality,which will increase the computational cost.The redundant attributes and irrelevant attributes also affect the detection effect.In order to improve the efficiency of network intrusion detection,this paper proposes two algorithms.The details are as follows:(1)Aiming at the feature selection problem of rough set theory,a feature selection algorithm based on information gain and rough set is proposed to reduce the time complexity.Firstly,the rough set theory is used to obtain the feature subsets,and then the information gain algorithm is used to select the majority and minority classes to obtain the feature subset with high information gain as the final result.Finally,a random forest classifier is used for classification.The experimental results show that the proposed algorithm based on the combination of information gain and rough set can not only effectively remove redundant and uncorrelated attributes,but also reduce the computational cost,and improve the classification accuracy and recall rate of minority classes without affecting the classification performance of majority classes.(2)Aiming at the feature selection problem of local linear embedding,a local linear embedding algorithm based on differential evolution is proposed to apply to the feature selection stage.For the parameter ? of?-neighborhood algorithm in local linear embedding,the proposed method uses differential evolution to optimize it.Using the optimized local linear embedding algorithm,the final feature subset is obtained from the original data by feature selection,and then the classifier is used for classification detection.The experimental results show that the feature selection method based on local linear embedding of differential evolution can eliminate redundant attributes,reduce space complexity and improve the performance of classification algorithm.