Research On Semi-Supervised Network Traffic Classification System Based On Deep Learning

In recent years,with the rapid development of the Internet,network traffic encryption has become an important means to protect network security.In order to further improve and optimize the network environment,how to accurately classify and identify network traffic has become an academic research hotspot.In order to optimize the classification performance of the semi-supervised network traffic classification system,this thesis propose a semi-supervised classification system based on Generative Adversarial Networks.In order to solve the problem of imbalanced dataset,Constraint based data enhancement algorithms are proposed.The main contents of this paper are as follows:(1)A semi-supervised network traffic classification system based on Generative Adversarial Networks is proposed.The classification model of semi-supervised network traffic classification module in the system considers the spatial and temporal characteristics of network traffic,cascades Recurrent Neural Networks and Convolutional Neural Networks to construct a discriminator network,and in order to prevent the model from overfitting through L1 regularization and Dropout.The length of best flow characterization is determined by simulation.The simulation results show that the classification accuracy of the proposed semi-supervised classification system is respectively 11%and 7%higher than that of the fully-supervised classification models 2D Convolutional Neural Networks and C4.5 when only 10%labeled data is used.Compared with the semi-supervised classification model based on Deep Convolutional Generative Adversarial Networks,this model is 5%higher for 10%labeled data,6%higher for 30%labeled data and 5%higher for 50%labeled data.In addition,The generalization ability of the proposed system is verified on UTSC TFC2016 traffic dataset,which is the malicious traffic dataset,the classification accuracy and F1 value on the three types of traffic dataset of UTSC TFC2016 all reached 0.99.(2)Constraint based data enhancement algorithms are proposed.In order to solve the problem of imbalanced dataset,the structural characteristics of the semi-supervised network traffic classification network proposed in(1)and the constraint of improved Tomek Link are used to increase the number of minority classes of sample traffic,so as to improve the classification performance of the classification model for a few classes of traffic.The simulation results show that the classification accuracy of data enhancement based on improved Tomek Link constraint is 1.32%higher than that without data balancing when the enhancement intensity of minority class is 10,that is,the ratio of minority class samples to majority class samples is approximately 1.00.To sum up,the proposed semi-supervised network traffic classification model based on Generative Adversarial Networks and Constraint based data enhancement algorithm can truly improve the accuracy of semi-supervised network traffic classification,and improve the precision and recall of minority network traffic.