Design And Implementation Of Network Traffic Classification System Based On Machine Learning

In recent years,the internet has become an important part of human social life.Network traffic management plays a vital role in guaranteeing regular operations of the internet and satisfying different requirements for quality of services.Accurate recognition and classification of network traffic provide technical guarantees for network management operation,such as traffic controlling,traffic monitoring,traffic statistics and quality of services guarantee.There are a variety of techniques can be used in network traffic identification,including port-based methods,methods based on network feature signature and methods based on network traffic statistic features.Machine learning(ML)classification algorithms firstly use the dataset to train a model,and then use the trained model to predict the characteristics of the target data.In this dissertation,some classical supervised ML algorithms’ performance in network traffic classification were compared,which provided the necessary reference for practical engineering application.The main work of the dissertation was as followings:1)The principles and characteristics of several classical ML classification algorithms were studied and analyzed.Based on the properties of network traffic data,analysis was made in how to apply these ML algorithms into network traffic identification and classification.2)A data model for online network traffic classification was proposed.Based on the ranking of information gain rates of attributes and the result of experiment,traffic features of the data model were selected.An evaluation of network classification models based on this data model was conducted.Verification the effectiveness of the data model in aspects of shortening the time costs of traffic characteristic processing and simplifying the structure of the model.3)The performance of three network classification models constructed from three different machine learning algorithms separately was analyzed and compared.Firstly,based on different sampling strategies to get training datasets,three different ML algorithms,that is Naive Bayesian,Support Vector Machine and C4.5 Decision tree,were used to train the classifcation models,and then the verifications of these models were carried out on specific test datasets.Lastly,based on the classification accuracy rate,recall rate,time consumption of training and verification and other related performance indicators,performances of three classification models in dealing with online network traffic classification was evaluated.4)A ML-based online network traffic classification prototype system based on B/S model was designed and developed.Built on the classification model with better efficiency,the prototype system can identify and classify 10 kinds of application categories.It consists of 3 modules,including network traffic monitoring,statistics of classification results visualizing,and specific traffic information displaying.