Bulakh V. Information technology of classification of ordered data arrays with fractal properties using machine learning methods

The thesis is devoted to the development of information technology for the classification of ordered data arrays (ODA), which have fractal properties, using machine learning methods. The methods for generating ODA with multifractal properties of various types are implemented in software. The series of experiments in which different types of ordered data split into classes according to their fractal properties were carried out. Ensemble methods of decision trees and neural networks were used as classifiers. The statistical, fractal and recurrence characteristics of ODA were used as features in the classification. Studies have shown that the range of multifractal and self-similar properties of data set plays an important role in the choice of a classifier and set of features, and, accordingly, classification accuracy. The proposed information technology analyzes the input information flow and selects classifier and features r to maximize the classification accuracy. The developed technology makes it possible to classify data with various fractal properties, for example, to detect DDoS attacks in infocommunication data, to clarify the diagnosis based on electroencephalogram and electrocardiography records, to classify seismic events for seismograms, etc.