Practical issues of the estimation and choice of machine learning models for the forecasts building in different subject domains

Abstract

The paper is devoted to the practical issues of machine learning models estimation for predicting processes in different subject domains. In the progress of work, there were undertaken the number of core steps. An analysis of theoretical and practical scientific sources on the use of traditional and modern models for forecasting in various domains is conducted to identify possible consequences of the use of different models risks. The features of building predictive models in selected domains (medicine, meteorology, finance, and sales) are determined. The criteria and their metrics for the models’ estimation are determined. To perform a comparative analysis and estimation of of machine learning models for forecasting processes in selected subject areas, a web application was developed. A number of predictive models are constructed with the help of the developed web application. The results of forecasting using traditional and modern models in the selected subject domains are analyzed and evaluated according to criteria of accuracy, speed and complexity. Based on the comparative analysis of of machine learning predictive models, the practical recommendations have been formulated for the correct choice of a model for specific domain forecasting tasks. The prospects of the research are outlined in the lines of automatizing the selection of better model.