Implementation of Classification Algorithms for Data Traffic Analysis

This article poses the task of analyzing network traffic using machine learning methods. The division of instant message service traffic into classes of possible user activities is made, the following classes are selected for markup: text, video, idle. Methods of mathematical statistics, machine learning are used, k-NN and Random Forest algorithms are compared, training is implemented by the Kohonen network. In the course of research on the example of the one of the information exchange applications, a method based on statistical analysis of network traffic, which does not require the study of the contents of the packages, was applied. As a result, the classifier is implemented, which can help to analyze data traffic. The efficiency of the classifier is demonstrated and the efficiency tested by cross-validation. The difference between the developed algorithms and the known ones is in the ability to effectively classify and track traffic without studying and decrypting the contents of packages.