Analysis And Classification Of Internet Traffic Based On Improved AP-SVM

With the use of a variety of new applications and network technology, the network environmenthas become increasingly complex, and it presents a huge challenge to network management. Inorder to better control and manage network traffic and guarantee the QoS of network application, itis necessary to classify network services effectively. In the traffic stream identification andclassification method based on statistical features,the key is the selection of features.This thesis analyzes tudou video, Skype, DOTA, QQ video and Thunder,which are the fivecommonly used services on Internet, based on the statistical features such as packet size distribution,Hellinger distance of the packet size distribution, packet size probability and transition probability,ratio of downstream and upstream packet number and bytes. The author finds that the packet sizedistributions of network applications are stable, and to some degree,the statistical characteristics ofthe average packet size distribution, such as variance, entropy, quartiles, skewness and kurtosis canreflect the traffic flow. Comparisons between the calculated traffic packet size distributionsHellinger distance find that the packet size distributions of tudou video and Thunder have someoverlap, while the packet size distributions of DOTAand Skype are very similar. However, from theperspective of the ratio of downstream and upstream packet number and bytes, these applicationsare easy to be identified. According to the analysis, the two characteristics: ratio of downstreamand upstream packet number and bytes, average packetsize can be used to classify the selectedapplications. Then, the existing AP-SVM algorithm is improved by proposing changablepreferences in clustering methods, getting a higher quality and more representative training sampleset, and better classification results. The experimental results of classification of network trafficshow that the improved algorithm can achieve better classification results.