Research On Network Traffic Classification Based On Deep Learning

Network traffic classification has always been one of the hot topics in academia,industry and network supervision.It refers to the classification of mixed traffic into different traffic categories according to the characteristics or parameters of different network applications or protocols.On one hand,the network security domain needs to identify the intrusion traffic;On the other hand,network management requires the classification and analysis of traffic in different applications,so as to reasonably control and allocate resources and ensure network QoS.With the increase of data volume and types of network traffic,the traditional classification method is difficult to meet the requirements,and the algorithm based on machine learning has become a research hotspot of network traffic classification.Aiming at the bottleneck caused by machine learning characteristic engineering,this paper studied the application of deep learning algorithm based on convolutional neural network in network traffic classification.The main work is as follows:First,in order to make full use of the time characteristics of network traffic data,this paper applies three-dimensional convolutional neural network to network traffic classification.After dividing the original network traffic into network flows,the same number of packets are extracted from the front of each flow,and the same length of each packet is retained.The sequence of packets in each stream is taken as a dimension,and the data in each packet is converted to two-dimensional by one-hot encoding.The pre-processed data is equivalent to the multi-frame grayscale graph in video processing and constitutes the input of the three-dimensional convolutional neural network.In this paper,the Tensorflow platform was used to build the three-dimensional convolutional neural network for simulation,and the effectiveness of the method was verified.Compared with the network traffic classifier based on the two-dimensional convolutional neural network,this paper achieved higher accuracy.Compared with the combined classifier which uses the recurrent neural network to process the time features,the number of parameters and the amount of calculation in this paper are obviously reduced under the condition of ensuring the accuracy.Secondly,the classification judgment layer of the convolutional neural network is improved for the errors caused by mistakenly dividing the unknown class into the known class.Through simulation experiments,this paper verifies that the distribution of probability value corresponding to the category with the largest probability is obviously different from the distribution of probability value when the category judgment is wrong,including the unknown category.Based on the above findings,a dynamic threshold is set for the category judgment layer in this paper.Under the optimal threshold discovered by the training process,this paper can effectively identify the unknown category.After dealing with unknown classes,the simulation results show that the classification accuracy of the system is improved by 11 percentage points.In addition,the unknown traffic data is marked and used to update the system,and the overall accuracy of the system is improved.