Encrypted Traffic Classification Based On Deep Learning

With the rapid development of society,the scale of the Internet is getting larger,and various types of network applications have emerged in an endless stream.The development of technology and various kinds of applications have met people's allround needs,but they have also brought huge challenges to network management and maintenance.Accurate and efficient classification of network traffic is a key step in network management.Due to the increased awareness of user privacy protection and the complexity of encryption algorithm,it is difficult to identify network traffic.However,traditional traffic classification methods can no longer meet the needs.Research and exploration of new traffic classification models and methods have strong practical significance and socio-economic value.Comparing with traditional models and methods,deep learning can build an end to end model,which has better classification accuracy and generalization,the paper studies the classification of encrypted traffic based on deep learning.The main research work includes:(1)Considering the timing characteristics of network traffic,an A-BiGRU encrypted traffic classification model based on attention mechanism is proposed.The model first encodes network traffic data from the input space to the feature space through embedding.Secondly,an encoder-decoder model is established,and encoder using BiGRU can better learning the context information of long network traffic sequences.The attention mechanism is used to enhance the model's ability to learn important features,then improve the classification accuracy of the model.Considering the imbalance of sample in the dataset,the focal loss function is used to improve the learning ablity of the class which have fewer sambles,thus improving the classification accuracy and generalization of the model.(2)The paper proposes a Dilated-1DCNN encrypted traffic classification model based on dilated convolution from the perspective of the model's performance in actual application deployment.The model optimizes the amount of calculation by constructing a point convolution module,and constructs a residual module to deepen the depth of the network model.And also,constructs a cascaded dilated convolution unit to increase the receptive field of the model,therefore,the model can obtain relatively high accuracy even dealing with long network traffic sequences.Then,the paper analyzes and calculates the complexity of the model,compared it with the A-BiGRU model,it shows that the amount of calculation is reduced by more than 10 times,which has significant performance advantages.(3)The paper conducts multiple controlled experiments using the “ISCX VPNnon VPN” dataset.The results show that the attention mechanism has significantly improved the classification accuracy of the A-BiGRU model;The focal loss function can effectively improve the learning ability of the model for the imbalance class which has fewer sample in the dataset.According to the experiments of the application classification and the traffic classification task,A-BiGRU model gets 0.99,0.97 F1-score,while Dilated-CNN gets 0.98,0.94.Compared with others models,the A-BiGRU model achieved a good precision results,and Dilated-1DCNN model get the least amount of parameters and calculation,which has a great advantage when aiming to deploy the model to engineering.