Research On Network Traffic Classification Based On Integrated Classification Algorithm

Network traffic classification technology for network security management has a crucial role.Due to the development and progress of network technology,the limitations of traditional network traffic classification methods such as port number recognition and classification method based on effective packet load detection technology have become more and more obvious,and can not classify the current traffic.Flow statistics network traffic approach based on machine learning has been the direction of today's network traffic classification technology as its lightweight and flexibility.In the method of machine learning,the supervised learning classification method and the unsupervised learning taxonomy have the shortcomings due to their own limitations.Therefore,the semi-supervising method uses the combination of supervised learning classification and unsupervised learning method to make up for the lack of two methods,which is very important to the development of network traffic classification method.This paper proposes a traffic classification algorithm based on semi-supervision.The Markov model is constructed by using the correlation information flows.The PCA method is used to extract the main feature components to reduce the dimension vector and eliminate the correlation between the features,and select the most effective method of four similarity calculation method for classification.Then designing the algorithm is used to select the initial center of the cluster,solved that the algorithm can not be optimized for the unknown traffic classification,and the accuracy is achieved.Finally,the clustering algorithm with the amount of different initial cluster which expressed in k is used to construct the integrated classifier,and the integrated classification result is selected by the Normalized Mutual Information as the optimal result.Experiments show that the classifier has achieved good accuracy,better solve the problem of port number recognition and the method based on effective packet load detection technology.