Research On Network Traffic Identification Based On Cluster Analysis

The increase of network traffic and the continuous development of network technology have brought great challenges to effective network security management and network traffic supervision.The premise of effective network security management and traffic supervision is the identification of network traffic.Therefore,how to identify network application traffic accurately and efficiently has become a hot research topic in today's computer network research.In this paper,the network traffic recognition technology based on clustering is studied and analyzed.In this paper,the method of semi-supervised learning network traffic identification based on k-means algorithm is studied.Because traditional k-means algorithm is easy to fall into the local optimal solution,which may resulting in the lack of clustering accuracy,the network traffic recognition technology based on traditional k-means algorithm can not obtain the ideal traffic recognition effect.The shortcomings of the K mean clustering algorithm are mainly caused by the inability to determine the optimal number of clustering and the random selection of the initial cluster center.Therefore,considering the two shortcomings of k-means clustering,this paper mainly improves the K mean clustering algorithm of semi supervised learning recognition method,and gets a better traffic identification method.By observing the distribution of network traffic,the improved algorithm selects the initial cluster center in the high-density area to improve the initial cluster center selection problem,at the same time,it introduces a clustering validity decision function to determine the optimal number of clusters,so as to improve the problem that the number of optimal clusters can not be determined.By comparing the network traffic recognition method based on the traditional K mean algorithm,it is proved that the improved algorithm has obvious advantages in recognition accuracy.This paper also studies the method of on-line network traffic identification.Most of the existing semi supervised learning method of network traffic identification needs to extract the statistical characteristics of complete flow,can only be used in the offline recognition,so this paper focuses on the research of online traffic identification,reference to the early-stage network traffic recognition technology and clustering technology this paper presents an online identification scheme.By referring to the early-stage network traffic identification technology,the scheme obtains characteristic attributes in the subflows which consist of the first few packets in the flow to solve the problem of online traffic identification cannot extract the required statistical characteristics.At the same time,the online traffic identification method is divided into training phase and online recognition stage.The initial clustering and mapping work is completed in advance at the training stage to solve the problem that the online incremental clustering cannot be mapped to specific network application types.Through further experiments,it is proved that the method has a good effect on processing performance and recognition accuracy.