Design And Implementation Of A Migration Learning-based Encrypted Traffic Classification System

With the rapid development of the Internet,the privacy and security of users due to frequent leakage of user data has received increasing public attention in recent years,and application vendors have started to prefer using encrypted traffic to give users a sense of security.However,due to the hidden nature of encrypted traffic,it also makes the traditional classification method based on plaintext traffic ineffective,and makes network automation and maintenance,network security management and network quality assurance services and other related network services bring great obstacles,the classification of encrypted traffic has become an urgent problem of network security and management.In recent years,transfer learning has made exciting performance in many fields such as natural language processing and computer vision,and also brings new opportunities for the development of encryption traffic classification.In view of the above problems,this thesis proposes a traffic classification system based on migration learning,and the main research results are as follows:(1)This thesis proposes a method of using transformer as feature extractor to classify encrypted traffic.Through comparative experiments,it is concluded that transformer is superior to CNN and RNN in feature extraction of time series information.The F1 value of 0.935 is achieved in the 12 classification task of iscx VPN non VPN,which is better than the existing methods.At the same time,easy assemble algorithm is used to solve the problem of data imbalance in iscx vpn-non VPN data set.(2)A classification method of encrypted traffic based on transfer learning is proposed.In view of the fact that there are too few public annotated data sets of encrypted traffic,which can not give full play to the performance of deep learning network,this thesis uses the migration based best pre training model to pre train the unlabeled data of migration learning on ctu-13 data set,and after the parameter pre training,it carries out fine tuning on iscx VPN non VPN data,and the F1 value of multi classification in the data set reaches 0.957.(3)An encrypted traffic classification system based on transfer learning is designed and implemented.Based on the market demand of detecting encrypted traffic types,five modules are designed and implemented,including front-end module,task scheduling module,data preprocessing module,traffic detection module and system storage module.Redis message queue is used to coordinate the task scheduling module for task scheduling,and traffic preprocessing and model classification are fully automated.In this thesis,the classification system of encrypted traffic will eventually be presented to users in the form of web pages to classify the traffic uploaded by users in real time.