Research On Network Traffic Classification Technology Based On Midway Identification

With the growth of the Internet scale and the increasing variety of applications,the social system has gradually become digital,networked and intelligent.Effective network traffic classification methods can improve the efficiency of network management and the accuracy of network analysis,thereby improving network security.In recent years,artificial intelligence technology has developed rapidly,and many researchers choose to use machine learning or deep learning in the study of network traffic classification methods.Although many existing studies have made great contributions to network traffic classification,there are still some problems.Most of the current network traffic classification methods are based on the complete protocol session flow,and cannot support network traffic identification after a session starts and lasts for a period of time,that is,midway identification.Such methods not only generate a lot of memory consumption and computational overhead in real scenarios,but also cannot well meet the needs of online realtime recognition in current complex scenarios.In response to the above problems,this paper starts from data preprocessing and model selection,and studies the traffic segmentation method suitable for midway identification scenarios.On this basis,two network traffic classification methods are designed and proposed.The main work is as follows:For data preprocessing,we first designs a time window-based traffic data segmentation method.This method divides the original traffic data into multiple traffic blocks by using a predefined time window as the basic data unit for subsequent processing.In order to realize the goal of identifying the network traffic in the middle,and based on this,two traffic classification schemes are designed,and in the experimental stage,time windows of different sizes are tested and analyzed to determine the optimal window size setting for the scheme performance.We designs a traffic classification scheme based on random forest.The solution is mainly divided into five modules: traffic collection,data preprocessing,feature extraction,model training and model evaluation.The data preprocessing module uses the above-mentioned traffic segmentation method to obtain multiple traffic blocks,and extracts the flow from the traffic blocks as features Extracted basic data unit.The feature extraction module considers three data packet sequences to extract statistical features based on the size of the data packets,and then normalizes the features.The model training module designs a multi-model voting mechanism based on the random forest algorithm,which improves the accuracy of model classification.Using the data set obtained by the traffic acquisition module,several sets of comparative experiments were designed,and the optimal parameter configuration of the scheme was determined.The results show that the average accuracy rate of this scheme reaches 89.23%,which proves the effectiveness of this scheme in the scenario of midway identification of network traffic.We designs a traffic classification scheme based on deep graph convolutional neural networks.In this paper,borrowing solutions from the field of graph classification,deep graph convolutional neural networks are used in the field of network security to solve the problem of network traffic classification.In the data preprocessing stage,the abovementioned traffic segmentation method is used to obtain multiple traffic blocks as the basic data units that are subsequently converted into traffic behavior graphs.The overall architecture of the traffic behavior graph including graph node definition,graph edge definition and node attribute definition is designed,and then the generated traffic behavior graph is input into a deep graph convolutional neural network for training.This model can retain more node information,and can learn global features from the topology of the graph;multiple sets of comparative experiments are designed to determine the optimal hyperparameter configuration.The experimental results show that the accuracy of this scheme reaches 93.69%,which proves the effectiveness of using the graph neural network to solve the problem of network traffic classification.Finally,comparing the performance of other schemes with other schemes in various evaluation indicators,the advantages of this scheme are illustrated.