Research On Network Traffic Classification Based On Semi-Supervised Learning

In recent years,as a basic technology of advanced network management and modern network security,network traffic classification has been widely used in network service quality control,intrusion detection and other fields.It allows operators and providers to effectively monitor and manage the network traffic.In today's networks,the advent of a large number of new applications make the composition of network traffic more complicated.The flow statistics-based traffic classification method can be combined with the machine-learning algorithm to realize the effective classification of network traffic by intelligently processing the data.Especially when combined with semi-supervised machine learning method,the extraction of unknown protocols can be realized.However,the existing network traffic classification method based on semi-supervised learning still has some defects in the practical application field.In view of these problems,this thesis has carried out the following research based on the original semi-supervised traffic classification method:In face of the complex network requirements,this thesis presents an adaptive multi-protocol classification system,which improves the traffic classification system based on semi-supervised learning from four aspects:1)The labeled flows are sufficiently utilized by providing information gain of each feature,confirming the initial clustering centers and mapping the traffic clusters to a certain protocol.2)Propose the method of dynamically adding the centers and iteratively calculating the semi-supervised k-means to achieve automatic parameter selection with high classification accuracy.3)Modify the cluster identification method to reduce the error mapping probability of unknown traffic clusters,and improve the extraction accuracy of unknown traffic.4)Optimize the system update program to improve the system's knowledge reserves and increase the system's classification of traffic types.A large number of simulation experiments show that based on these improvements,the system can effectively extract the unknown protocols while accurately classifying the traffic of many common protocols in the network.On this basis,this thesis also proposes to combine the improved adaptive semi-supervised traffic classification method with load balancing technology.Feasibility and advancement of this combination are analyzed from three aspects,which makes the load balancing system provide more intelligent and diversified services.