Research On Clustering Analysis Based Application Layer Traffic Classification

With the rapid development of Internet service in recent years, many new applications,represented by P2P,occupy a large number of network bandwidth and reduce network performance, furthermore,they bring great risks to network security. Therefore,traffic classification is a hot topic of research. It is very important to identify application traffic for network management, network security and planning etc.The thesis studies traffic classification using clustering analysis. Firstly, we summarize and analyze the existing techniques about traffic classification, and then we propose two modified algorithms,which are the traffic classification algorithm based on improved K-means and the traffic classification algorithm based on genetic clustering algorithm.The experimental results show that these methods are valid. The main results of this thesis are as follows:A traffic classification algorithm based on improved K-means is proposed.For the deficiencies of the K-means algorithm:firstly,the clustering result fluctuates according to the random selection of initial centers.Secondly, it considers all attributes of flows have the same influence on similarity.Some traffic like P2P can distinguish from others by attributes,for example, counts of transmission bytes and continuous time,so it can reduce the accuracy.And then, there exists of a large number of relevant attributes for traffic matrix,which affect the clustering efficiency. In order solve these problems, a traffic classification algorithm based on improved K-means is proposed. Firstly,it proposes an initial center of optimization method.Secondly, it introduced Singular Value Decomposition technology to simplify Matrix.And then it assigns weight to attributes,which represents the influence of each attribute.The experiments show that it can achieve better performance compared to the genetic clustering algorithm, it also improves the accuracy of traffic classification.A traffic classification algorithm based on genetic clustering algorithm. For K-means's requirement about number of clusters k in advance,and random selection will lead to a decline in clustering quality, a traffic classification algorithm based on genetic clustering algorithm is proposed. This method puts the range of clustering number k as the search space,and constructs fitness function using improved K-means.It can automatically find the best clustering number of categories, and achieve better performance of traffic classification. At last,the paper extracts flows from dataset to form flow matrix by Winpcap library.It establishes a traffic identification platform for the traffic classification algorithm based on improved K-means and the traffic classification algorithm based on genetic clustering algorithm.The experimental results indicate that the two methods can not only convey that the method can obtain high classification accuracy and solve traffic identification problems effectively compared to the previous works, which also can reduce the time complexity, it also improves the quality of service about traffic classification.