Network Video Flow Classification Model Fusion Based On Similarity Measurement

Currently,Internet technology is developing rapidly,Emerging applications came into being.With the complexity of the network environment,Internet Service Providers(ISP)urgently need to classify and manage a large number of network video streams.At the same time,the development of network technology has caused a large number of previously collected and labeled network flow data sets to become obsolete,resulting in fewer training sets available,and a large amount of new data needs to be collected and labeled in real time;In addition,network flow features will experience conceptual drift with time and network environment changes,and the drift of flows of different application categories may be different,resulting in reduced accuracy of flow classification methods based on traditional machine learning(ML)models.To solve the above problems,this paper proposes a network flow fusion model based on similarity measurement,and also proposes a hybrid feature selection(FS)method SUmax-GA to improve the classification accuracy.This article focuses on asymmetric video streaming: on-demand ultra-definition video(CD/1080p),on-demand high-definition video(HD/720p),on-demand standard-definition video(SD/480p),and symmetrical video streaming: instant messaging video(IVC),P2 P video(P2P)and online live video(ILV)are classified into six video services.The main research work is as follows:Combining SUmax and genetic algorithm(GA),a hybrid feature selection(FS)algorithm SUmax-GA is proposed.It combines the advantages of filtering and encapsulation FS methods to achieve the reduction of feature dimension and feature redundancy.The algorithm first uses the SUmax method to sort all the feature correlations,filters out some irrelevant features,and uses the remaining features as the initial population of GA.Next,the accuracy of the CART classifier is used as the adaptability evaluation function to evaluate the pros and cons of the feature subset,and the best feature subset is selected through continuous adjustment and optimization based on the GA packaged FS method.The experimental results show that the optimal features selected by SUmax-GA can effectively improve the performance of the classifier.Based on JS distance(Jensen-Shannon distance),Multi Tr Ada Boost and RF(Random Forest)methods,a similarity measure-based network flow classification fusion model JSD-MTAB-RF is proposed.The fusion model uses JS distance to measure the similarity of the distribution of new and old data sets,combined with ML and TL,can make better use of past data,while saving the cost of labeling new data,the accuracy of classification is improved.Compared with Multi Tr Ada Boost in [26],the classification fusion model method in this paper can achieve better results.Using the FS method SUmax-GA proposed in this paper and the existing FS method MSGA and the fusion model JSD-MTAB-RF in the literature [26],experiments on the actual network traffic data set show that the method in this paper can achieve more than 96% The classification accuracy rate is improved compared with the existing MSGA method;Using the classification fusion model proposed in this paper to compare with the classic multi-class transfer learning algorithm Multi Tr Ada Boost,the experimental results show that the method in this paper can achieve a classification accuracy of more than 95%,which is better than the classic multi-class transfer learning algorithm.