Research On Feature Subset Selection Based On PCA And It's Application On Network Intrusion Detection

With the development of networks,the security of network information has become increasingly prominent.Due to the increase in the amount of data in the network,the difficulty of judging intrusion data continues to increase,so an efficient intrusion detection technology is urgently needed to ensure network security.In recent years,machine learning has been widely used in many fields.Research has shown that machine learning has good results in the field of intrusion detection.However,machine learning algorithms rely heavily on data.The large number of network data features and large redundancy will affect the effect of intrusion detection.Therefore,the feature selection for network intrusion detection is of great research value.Principal Components Analysis(PCA)can reduce the dimensionality of features,but this method cannot effectively eliminate the influence of irrelevant features,and the result after dimensionality reduction loses the actual physical meaning,which is not conducive to the study of important features of intrusion detection data.This paper proposes a feature selection method combining PCA and improved genetic algorithm.First,PCA is used to reduce the dimensionality of the intrusion data,and calculate the cosine similarity between the principal component and the original feature and get the feature group;the same grouping feature is used as the similar feature,and the genetic algorithm is used to extract the grouping features.In order to improve the algorithm's ability to optimize the grouping features,each time the genes in different groups are selected for crossover and mutation,and the detection performance of the classifier is used as the fitness function of the genetic algorithm,and the optimal feature subset is selected as the intrusion detection data feature.This paper uses the KDDCUP99 to conduct experiments,and combines the feature selection method of this paper with different machine learning models for intrusion detection experiments.Results show that the feature selection method in this paper reduces the number of features and ensuring the classification performance of the model.