New Voting Combination Classification Method For Network Multimedia Services

With the rapid development of Internet technology,various types of network services continue to emerge.Among them,the network multimedia service consisting of network video services,network browsing,network music,and network games is an increasing part.The classification of network multimedia services helps network service providers(ISPs)to manage network resources and provide personalized Qo S guarantees.This thesis classifies four types of online multimedia services: web browsing(Web),online live video(CBox),online non-live video(Youku),and online games(LOL).The main research works are as follows:Performing data collection of four typical network multimedia services,and a modified feature extraction method is designed for feature extraction.A visualization method is used to judge the validity of the feature set for flow classification.Comparative experimental results show that the feature set extracted by the modified feature extraction method has better result.Based on the idea of ensemble learning in machine learning,the new voting combination classification method for network multimedia services is proposed.This method first selects three base feature selection methods based on the distribution of different data sets.The features selected by the three base feature selection methods are classified,and the classification results are combined to obtain the final prediction label.In the process of voting combination,this thesis introduces the indicator of confidence,which can improve the accuracy of the final predicted label when the three initial predicted labels are different.This thesis carried out many comparative experiments on the UCI dataset and the network multimedia service dataset.Combined experiments for different base feature selection methods.Comparative experiments on different machine learning classifiers.Contrast experiments for the number of different features.Experiments show that compared with the existing methods,our method can improve the classification accuracy and has better adaptability to different data sets.