Research On Intrusion Detection Technology Based On Multi-Feature Fusion And Ensemble Learning

The rapid development of information technology has made network security issues more and more indispensable.Regardless of country,enterprise or individual,the leakage of important information will cause serious losses.In order to maintain network security and prevent information leakage,intrusion detection as a security protection technology has become a means to ensure information security.With the rise of artificial intelligence,intrusion detection has become intelligent,so machine learning is applied to intrusion detection,and the intrusion detection system is more intelligent under the premise of improving accuracy.This article focuses on systematic research on intrusion detection based on two methods: feature selection and integrated learning based on multi-feature fusion.First,pre-process the two intrusion detection data sets.Analysis of the NSL-KDD data set.Because this data set was collected earlier,the UNSW＿NB15 data set is used.The attack type of this data set is more in line with the intrusion type of modern networks.Pre-process the above data sets separately to obtain data that can be used for the next experimentSecondly,intrusion detection methods and experimental schemes based on multi-feature fusion are designed and implemented simultaneously.Due to the high dimensionality of the collected data sets,the efficiency of machine learning processing is low,and a feature selection method of multi-feature fusion of principal component analysis combined with linear discriminant analysis is proposed to remove the feature attributes with small impact and class separation.Afterwards,k-proximity algorithm and naive Bayes algorithm are used to classify the two data sets respectively.Experiments show that the data detection rate after multi-feature fusion processing is improved to a certain extent,and the calculation time and running space are saved.Finally,an intrusion detection algorithm and experiment based on integrated learning are designed and completed.The KNN algorithm is used as the base classifier,and the Bagging algorithm is used for integration;the decision tree is used as the classifier,and the Adaboost algorithm is used for integration.The two integration methods are applied to intrusion detection technology,and experiments are conducted on both intrusion detection data sets.The results show that the integration algorithm is compared with the original weak classifier to improve the efficiency of intrusion detection.