Application of machine learning algorithms in determining the value perspectives of corporations

Abstract

The article solves an important problem of effective application of machine learning algorithms in the process of determining the perspectives of corporate value. The results obtained will allow reducing the losses incurred by companies as a result of value fluctuations by preparing in advance. The investigation employed the subsequent approaches: model modifications, automatic search of coefficients, construction of several models with different cut-off dates, support vector regression, etc. The conceptualization of the methodological approach to model modification based on the exclusion of suboptimal models and the comparison of model residuals and white noise is developed. The suggested approach comprises the subsequent phases: analysis and pre-processing of the data set; division of the prepared data into training and test samples; modeling and forecasting based on the modified model and the error limitation model; evaluation of the results. The architecture of an information system for forecasting based on time series models is developed. The efficiency of building multiple forecasts for solving machine learning problems is investigated. The substantiated recommendations will help to increase the accuracy of forecasting the perspectives of the value of corporations within a certain confidence interval.