Research And Implementation Of Traffic Classification Platform Based On Machine Learning

With the increasing amount and data types of network traffic,traditional network traffic classification methods have been difficult to meet the demand.The network traffic classification method based on machine learning has become a research hot spot.However,development of machine learning models is a cyclic process and requires multiple optimizations.In practical applications,development of network traffic classification models based on machine learning involves many aspects.Traditional development methods are inefficient.Now,there are many open machine learning platforms.But we have not found commercial machine learning platform for network traffic classification.So,we designed and implemented a traffic classification platform based on machine learning.In this thesis,through the organic combination of machine learning technology and network traffic classification field,we built a services network traffic classification platform.It contains data set management,model development,model training,model evaluation and model online prediction application.According to the characteristics of the platform,the entire platform contains six modules,Homepage Module,Data Exploration Module,Data Mining Module,Data Prediction Application Module,User Management and Personal Account Management Module.Data Exploration Module realizes the storage management and query functions of the network traffic data set for training and prediction.At the same time,to achieve the efficiency of selecting features and improving the training quality of the model,this module also realizes the visual display and feature engineering functions of the stored data set of the platform.Data Mining Module contains two submodules of Task Management and Model Management.They achieve effective implementation of algorithm model import,training model task scheduling,and training task process management.Data Prediction Application Module implements the evaluation function of the developed model.Besides,it performs unified storage management on the completed model to achieve model persistence.It is convenient for the prediction application of the model in the future.In addition,before the design and implementation of the platform,we studied network traffic classification based on machine learning.We also collected network traffic data,explained network traffic data preprocessing and feature extraction and selection.We selected and implemented a network traffic classification model based on Naive Bayes and SVM classification algorithms.The final model has been provided to the platform for commercial use.This platform is designed using B/S architecture,which is realized by separating front and back ends.It adopts a component-based development method and uses new design schemes such as dva data flow management and React-based Ant Design Pro for development.The traffic classification platform based on machine learning is described in this thesis.It is in the stage of development and improvement.The test version has been released for internal employees of a unit.Through platform management,it greatly improves the development efficiency.Test shows the platform has good engineering practicability.