Research And Implementation Of Network Traffic Classification System Based On Fuzzy K-Means Algorithm

With the continuous development of network, especially the mobile Internet and Internet of things, the network has become an indispensable part of life. Along with the development of network, network security, quality of service, network management and some other related issues have become increasingly prominent.If we can not effectively manage the network traffic, we will have a significant impact on our daily production and life. The basis of network traffic management is to identify network flow classification effectively, however, the effectiveness of network flow classification method based on the port or based on the load has been greatly weakened. Many researchers have begun to turn their attention to machine based learning.Therefore, aiming at the fuzziness of the network flow, this paper studies the application and effect of the fuzzy clustering algorithm in the classification of the network flow.This paper firstly compares the network flow identification method based on port, load and machine learning.Then introduces the concept of network flow and evaluation criteria, and points out the advantages of these three methods, inadequate, the use of the scene.Then, the selection method of network flow statistics is discussed, and an improved fuzzy clustering algorithm is proposed for the classification of the network flow, and it is applied to the identification of the network flow.Finally, a classification system of the network flow is realized.The main contents of this paper are as follows:(1) In the selection of network flow characteristics, this paper considers two aspects:the artificial experience selection and the machine based learning, and combines the advantages of the two to find the appropriate collection of network flow, as much as possible to consider related features to improve the classification accuracy, while maintaining acceptable computation.In this method, the influence of the different feature groups on the network flow classification is investigated by the artificial experience firstly.Then gradually refine, to find out which features of the network flow classification has a relatively high degree of contribution.Finally, through the machine learning algorithm to compare the classification results of the selected features, and determine the final appropriate collection of network flow.(2) According to the fuzziness of the characteristics of the network flow, this paper proposes an improved fuzzy K-Means clustering algorithm.This algorithm is used to describe the relationship between the flow characteristics and the different application protocols by fuzzy weight.In view of the impact of the classification accuracy of the algorithm is easy to be affected by the choice of the initial cluster center, this paper improves the algorithm based on the minimum spanning tree algorithm.Finally, the improved fuzzy K-Means clustering algorithm is applied to the network flow classification.Design and implementation of a network flow classification system, the system can complete the capture of network data packets, filtering, aggregation, feature extraction, classification and output display and a series of operations.Finally, the system is tested on several kinds of data streams which are captured by the system.The experimental results show that compared with the traditional K-Means algorithm based on the improved fuzzy K-Means algorithm, the accuracy of the network flow classification is improved.