Campus Network Game Traffic Identification Based On End-to-End Strategy

With the rapid development of campus informatization,network security has become a focus topic of close attention in colleges and universities today.Among them,network attack and malicious traffic detection,high-performance network protocol design,network operation management,network development planning and other management requirements are increasing.The network traffic classification as the technical basis for the realization of the above requirements is both a theoretical basis and an important technical means.Online game traffic occupies campus network resources also affects students' learning,and refined management requires precise control based on the proportion of online games occupying resources and the impact on students' learning.Although online game traffic is a type of network traffic,it is quite different from the traffic of other network applications.Part of the online game traffic uses tunnel encapsulation technology,using encryption protocols for encryption,and accounts for a very small proportion of the public traffic data set.The existing network traffic classification technology cannot achieve accurate and comprehensive classification.The annotation of online game traffic depends on the implementation of anti DNS resolution,while the maintenance of DNS list costs a lot of labor and time,so it is still difficult to annotate a large number of network traffic data sets.This paper proposes a game traffic annotation method based on port mapping,and uses a characterization learning algorithm under the end-to-end strategy as a classification model to optimize the classification effect of the traffic classification model by improving the data set structure,better identify the network game traffic,and proposes a new model input structure,which effectively improves the model training efficiency on the premise of ensuring a certain recognition rate.In this paper,feature engineering is used to deal with online game traffic,which is extended to open data set,decision tree algorithm is used to classify online game traffic,and the feasibility of game traffic tagging method is verified by comparing the classification effect of data set before and after expansion.Then the structure of the data set is optimized by the analysis of the experimental results in the last stage,and the different performance of the data set under the representation learning algorithm before and after the optimization is compared,which provides the experimental basis for the further optimization of the model training.Furthermore,a 1* m*n model input structure is proposed,which can reduce the sample dimension and improve the computational efficiency while improving the classification accuracy of the model.Finally,compared with the experimental results under different algorithm models,it is proved that the two-dimensional data set construction method proposed in this paper has stronger stability and generality for characterization learning,which makes the classification of online game traffic better than that of general traffic.The accuracy of the classification model was increased from 81% to 97% and from 92% to 98%.