Research On Network Traffic Classification Method Based On Machine Learning

With the rapid development of Internet technology,various network applications have provided great convenience to people's lives,the number of network access users continues to increase,the data volume of network traffic is growing exponentially,and there are more and more potential threats in the network environment.In the field of network security and network operation services,the accurate classification of network traffic can realize the perception of abnormal conditions in the network environment,thereby improving the security of the network and the rationality of resource allocation.In order to solve the problem of the difficulty of feature selection in traffic classification by machine learning methods,this thesis conducts an in-depth study on the principles of convolutional neural networks,and designs an end-to-end traffic classification method based on convolutional neural networks.In this thesis,the original traffic coding form is transformed into a matrix form of hexadecimal,and each value in the matrix corresponds to a pixel in the gray image.Thus,the traffic classification problem is transformed into an image classification problem,and the process of feature selection is omitted.Then the structure and super parameters of lenet-5 network are adjusted to meet the classification requirements of normal traffic and malicious traffic,and the effectiveness of convolutional neural network for traffic classification after image processing is verified.On this basis,for the problems of deep convolutional neural network with many parameters and high computational complexity,a feature fusion network based on the bottleneck layer and the Inception network module is proposed.For the input layer data,use parallel double-branch convolution operation for feature extraction,and fuse the feature maps obtained from different receptive fields to obtain richer features.The introduction of the Inception module replaces the convolutional layer stacking method,expands the network in width,improves the generalization ability of the model,and before input to the large-size convolution kernel,the bottleneck layer is used to reduce the dimension of the feature map channel,which effectively reduces the parameter quantity and the over fitting.The experimental results show that for the original network traffic data after image processing,the improved Le Net-5 network is used for classification,which can effectively improve the accuracy of traffic classification.The classification accuracy of lightweight feature fusion convolutional neural network for encrypted traffic is more than 95.5%,which is1.8% higher than the improved Le Net-5 network,and the classification accuracy in different classification tasks has reached more than 90%,which proves the effectiveness of the improved method.