Dissertation > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Computer network > General issues > Computer networks, test , run

Research of Network Traffic Classification Based on PSO Optimized Semi-supervised Learning

Author ZhangLongZuo
Tutor LiuBin
School Huazhong University of Science and Technology
Course Information Security
Keywords Flow characteristics choice PSOKNN PSOKmeans Semi-supervised machinelearning Session hijacking
CLC TP393.06
Type Master's thesis
Year 2013
Downloads 3
Quotes 0
Download Dissertation

In the high-speed development of the Internet, along with the emergence of P2Pdownloading and P2P streaming media technology, many kinds of new networkapplications have sprung up. The sharp increase in the network load makes traditional ISPcome to realize that the Internet situation is grim. How to identify the various types ofnetwork applications quickly and effectively in the high-speed flow of network bandwidthand feed back the efficient recognition results to the actual network control, and thus bettercontribute to the management and control of network traffic flow detection have becomefield problems to be solved.In such situation, the work commenced mainly from thefollowing aspects:Selecting flow characteristics as the characteristics of the network traffic classification,an efficient flow feature vector selection method is proposed. The method is not only tocover the traditional network applications, and deliberately added to the flow eigenvectorswith higher P2P streaming media discrimination. The selection of feature vector is morecomprehensive and more differentiated.For the lack of unsupervised learning method such as Kmeans and KNN, proposed aoptimize algorithm using particle swarm optimization technology. In response to thetime-consuming problem that labeling samples for supervised learning characteristics isdifficult, introduced the concept of semi-supervised machine learning. Ultimately designedand implemented semi-supervised learning network traffic classification systems based onparticle swarm optimization. The classification system includes real time trafficclassification and offline traffic classification, the real time traffic classification system ismainly responsible for the online real time network traffic monitoring. The design ofclassification system has high real time requirements, so in the process of systemimplementation, a model based on average packet size is proposed. Such a model caneffectively deal with the real time traffic classification. Offline traffic classification systemresponsible for the offline distinction of network traffic, mainly focused on the fine work ofthe classification accuracy and the discovery of new applications.According to the real time preliminary classification and refined measurement of offline traffic classification system, many campus network users need to be warned. For thefreedom of campus network, this paper proposed and designed an alarm system based onthe session hijacking. This system is simple and easy to deploy. It could send timelywarning information to the campus network users to ultimately achieve the purpose ofmaintaining the stability of the campus network traffic.

Related Dissertations
More Dissertations