Research On BiLSTM-BahAttention Encrypted Traffic Recognition Combined With COUT Algorithm

With the advancement of network information transmission technology,the current world is in a situation of explosive growth of information,and people have higher and higher requirements for the security of sending or receiving information.In order to improve the security of network information,encryption technology is widely used in the information transmission process.At present,most traffic is transmitted securely in the form of cipher text.Traditional traffic identification technology is difficult to identify encrypted traffic efficiently.For users and service providers,it is necessary to propose an effective identification method for encrypted traffic,which is an important prerequisite for ensuring network security and ensuring service quality.Network users use different applications at different frequencies,so the data sets obtained or used usually have uneven samples.In order to solve this problem,the COUT algorithm is proposed.The COUT algorithm is guided by the K-means algorithm and combines oversampling and undersampling techniques.The COUT algorithm not only solves the problem that the SMOTE algorithm is prone to noise points,but also solves the problem of blindly deleting samples in the random undersampling algorithm.Through four sets of experiments,the effects of COUT algorithm,SMOTE algorithm and random undersampling algorithm on the sample recognition accuracy were compared.The results show that the COUT algorithm can improve the model recognition accuracy and the improvement effect is better than other algorithms.Aiming at the problem that traditional traffic identification methods cannot effectively identify encrypted traffic,a BiLSTM-BahAttention network model is proposed.Based on the BiLSTM model,the model replaces the Sigmoid activation function and Tanh activation function in the original model with the Leaky Re LU activation function.The convergence speed of the model is improved,and the problem of gradient disappearance is avoided;and the attention mechanism Bahdanau Attention is introduced,which further improves the feature extraction ability of the model.In order to test the ability of the BiLSTM-BahAttention network model to identify encrypted traffic,this paper selects the ISCX VPN-Non VPN 2016 data set.First,the original data set is preprocessed through sample denoising,sample length normalization and COUT algorithm,and then the experimental samples are passed through The BiLSTM-BahAttention network model is used for identification.Finally,the F1-Score and Accuracy obtained in the VPN six-classification experiment and the VPN-Non VPN twelve-classification experiment are better than the network models CNN,RNN and BiLSTM.