Machine Learning Methods as an Alternative to Factorization of Multidimensional Data

Abstract

Multidimensional random data and information flows often have different or limited numerical dimensions. When analyzing the intra-system relationships of such data, correlation and factor analysis are ineffective. The purpose of the study is to evaluate the possibilities of the combined use of the “decision tree” method and the artificial neural networks for the analysis of multidimensional random data. Materials and methods. Machine learning methods are used to classify multidimensional random data with different numerical dimensions and statistical distribution. The analytical platform “Deductor” is used as the software. The experimental data set contains 27 random parameters. The system analysis was carried out on a sample of 200 to 500 values of each parameter. Results. It is shown that the proposed approach to the system analysis of multidimensional information flows has a number of advantages over traditional correlation and factor analysis. It does not impose restrictions on statistical distributions, allows one to work with a limited data sample, and select the most significant parameters. Conclusions. The combined use of machine learning methods allows one to significantly reduce the training sample without losing the calculations accuracy. For technical applications, this makes it possible to receive and analyze information dynamically, in real time using standard microprocessor equipment. The results can be applied in the tasks of information exchange and cybersecurity of the electric power industry.