| Network traffic classification is of great significance for network management,network security and network optimization.With the development of network technology,the traditional network traffic classification model has been unable to meet the needs of the present.The traditional network traffic classification method has many shortcomings,such as being restricted to dynamic port,data encryption,low classification accuracy and high computational cost.This paper presents a network traffic classification method based on PCA feature selection and optimized ECOC,combined with the SVM algorithm,which is used in actual network traffic classification.Aiming at the problems of too much traffic features,features redundancy and high computational cost,this paper proposes a novel feature selection method.On the foundation of principal component analysis of network traffic,the threshold of cumulative contribution rate is set,and the network traffic feature is selected through finding the components with the biggest mangnitude in eigenvalue matrix.Thus,the final classification features can bu reduced to about 25% of the original features,and achieves the purpose of feature selection,data dimension reduction and computational cost reduction.The comparison experiment shows that the proposed method takes effect in improving classification accuracy and reducing the overall computational cost.Aiming at the problems existing in the traditional ECOC classification method,such as coding efficiency is not high,samples in sub classifier are unbalanced and the encoding matrix selection is usually unreasonable.This paper optimizes the traditional ECOC method,proposes a comprehensive method of constructing Hadamard matrix,and specifies the selection principle of encoding matrix.Then,we obtain the final encoding matrix with more tight order and better adaptability for classification number,and achieves the purpose of ECOC encoding table reconstruction.The optimized ECOC classification method has higher coding efficiency,and overcomes the problem of sample unbalance.The experiment result shows that compared with the traditional ECOC method,the proposed method has higher classification accuracy.In this paper,the experimental data is selected from actual network traffic data ‘Moore set',which reflects the real network environment.So the experiment result has certain universality. |
