Research On Trusted Encrypted Network Traffic Classification Metho

Network traffic classification refers to the classification and identification of collected network traffic data of various applications,which is widely used in research fields such as network resource allocation,traffic scheduling,and intrusion detection systems.With the widespread application of encryption technology in network communications,the classification of encrypted network traffic has become a hot research topic.At present,most of the existing methods only focus on the accuracy of encrypted network traffic classification.However,there is very little research on the trustworthy of the classification model.In this thesis,a trustworthy encrypted network traffic classification method is proposed,which has high classification accuracy and effectively corrects the confidence of the model output to improve the trustworthy of the classification model.Specifically,the proposed model is mainly divided into two parts.Firstly,in the data preprocessing stage,the original network traffic data is segmented in the form of flows and sessions respectively,and then the first 784 bytes of data of all layers are extracted and converted into image format.In the trustworthy traffic classification stage,the preprocessed original network traffic data is first learned through convolutional network layer,and then the learned features are sent to the classification network for initial encrypted network traffic classification.Meanwhile,a confidence network is designed based on the features extracted from convolutional network layer,and use the probability of the true class to train the confidence network to output a credible confidence value.According to the confidence value,the correct prediction and the wrong prediction can be effectively distinguished,thereby improving the trustworthy of the model.In this thesis,the experimental performance of the algorithm is verified on two publicly available encrypted network traffic datasets ISCX VPN-non VPN and USTCTFC2016.In terms of classification accuracy,the ISCX VPN-non VPN data set was used to extract flow and session data and perform classification experiments.The results show that the classification accuracy of the proposed method can reach 87%.In terms of misclassification detection,the comparison with the existing model proves that the proposed model shows superior performance in misclassification detection.On the session data extracted from the USTC-TFC2016 data set,the proposed model is on the AUROC indicator reach to 97.77%.Finally,by analyzing the confidence values of specific samples,it is verified that the proposed model can effectively detect misclassified traffic samples.