Feature Selection And Classification Of Internet Video Traffic Based On Genetic Search Algorithm

In recent years,as the characteristics of openness,connectivity,and sharing of networks have become increasingly prominent,in order to meet the increasing needs of people,new types of network applications and various application layer protocols are emerging,making the network traffic environment more complex..How to reasonably allocate network bandwidth resources and ensure QoS(Quality of Service)of multimedia services in a complex network environment has become an urgent problem to be solved by an ISP(Internet Service Provider).Network traffic classification allows ISPs to provide corresponding levels of service based on different types of video applications,which in turn makes network resource allocation more reasonable.This paper mainly categorizes online video,instant messaging video(ie,QQ video),P2 P video(kankan video),and HTTP video.Online video includes live streaming and online non-live video.Online non-live video includes online SD,online HD and online ultra-clear video.This paper proposes a novel feature selection method based on genetic algorithm and symmetric uncertainty.To verify the effectiveness of the method,a classification experiment is combined with the statistical characteristics of seven network video service flows.Based on three feature selection search algorithms: genetic algorithm,greedy search algorithm and particle swarm optimization algorithm,another evaluation function is compared,ie,inconsistent rate,and feature selection.Then using BayesNet,J48 and RandomForest three classifiers to classify seven kinds of video service flow classification experiments,and finally from the three aspects of accuracy,F measure and time complexity comparative analysis,verify the effectiveness of the method.In order to filter out useful data from a large amount of data,the initial feature set can be preprocessed,ie,discretized.This paper compares the classification accuracy and F measure before and after feature discretization,and verifies that the discretization operation can improve the average classification accuracy and F measure of video service flow,and can reduce the time complexity effectively.It is difficult to accurately distinguish different video services based on the distribution of a single feature.How to find a combination of features with a high degree of differentiation is crucial to the classification of network traffic.Through the method proposed in this paper,we can choose the best combination of features.