Research And Implementation Of Encrypted Traffic Recognition Based On Machine Learning

With the gradual improvement of China's Internet information infrastructure,the scale of China's Internet traffic has also increased year by year,which has greatly promoted the progress and development of our society,but has also caused hidden dangers in network security incidents.For the security of network information transmission,more and more traffic data is now gradually being encrypted.However,with the upgrade of some attack methods,many malicious traffic attacks are also carried out through encryption.This gives traffic Identification brings new challenges and new hidden dangers to network security.Therefore,this paper proposes an encrypted traffic recognition scheme based on machine learning,and implements a set of traffic visualization and detection system.The main research results and innovations are as follows:Aiming at the problem of time-consuming operation of the traffic recognition model caused by many encrypted traffic features and high dimensions,this paper proposes to filter features by establishing a feature library and feature selection,and use the features after feature selection to train the recognition model to complete the encrypted traffic.classification.This paper uses about 250,000 public data sets,uses the machine learning framework XGBoost to train and test the model,and finally achieves xx accuracy.The experiments in this paper verify that the method can significantly improve the detection efficiency with a limited loss of accuracy.Aiming at the problem that the neural network model has many parameters and a large amount of calculation,this paper designs an end-to-end distillation network model YOTO-Net to reduce the weight of the traffic recognition model.The end-to-end distillation structure design reduces the waste of training resources.In addition,a multi-task loss function is designed in this paper.The loss function not only becomes smooth but also speeds up the convergence of the network.This paper was trained and verified on Pytorch,and finally achieved an accuracy of xx,which proved the effectiveness of the research results.Combining the above two research results,this article designed a set of encrypted traffic online analysis system,integrating data acquisition,data preprocessing,encrypted traffic identification,data visualization and other modules.Each module works independently through decoupling design to ensure that The stability and scalability of the system are improved.In actual tests,the system performs well and has excellent performance.