Problem of Network Traffic Classification in Multiprovider Cloud Infrastructures Based on Machine Learning Methods

摘要

The development of new services, VoIP, video calls, multimedia streaming, streaming of high-quality content, and the Internet of Things (IoT) technology is forcing communication providers to rebuild cloud infrastructures in accordance with new requirements. Analysis of traffic data allows effectively reconfiguring the structure of a multiprovider network, taking into account the user's network portrait. In first part of this work an overview of the tasks solved in multiprovider cloud infrastructures by the classification of network traffic is showed. Then authors describe classical machine learning algorithms (Naive Bayes classifier, SVM, kNN, decision tree) that can be used to solve the assigned tasks, and revealed their advantages and disadvantages. In experimental part the problem of traffic classification of various types of applications generated by the D-ITG tool is decided, and a comparative analysis of machine learning algorithms is carried out. The main comparison metrics are precision, recall, F1, accuracy, complexity and time of training and testing. According obtained results decision tree method is chosen as optimal method that shows the maximum accuracy and fastest time of learning.