USAGE OF DATA MINING METHODS FOR FORECASTING FOREIGN ECONOMIC ACTIVITY INDICATORS

Abstract

The article describes the results of the study of machine learning algorithms usage for foreign economic transactions in Ukraine analysis and forecasting. The purpose of this article is to forecast import and export indicators using machine learning algorithms (linear regression, Gaussian Process Regression, SMOreg and Multilayer Perceptron neural network) based on statistics covering the period from January 1, 2018 to December 31, 2021. The most accurate forecast result was identified by using statistics for different intervals of the base period and forecast periods. The accuracy of machine learning algorithms was assessed by comparing the following indicators: mean absolute error (MAE), root mean square error (RMSE), and mean absolute error in percent (MAPE). The calculated forecast indicators of foreign economic operations according to the SMOreg algorithm have high forecast accuracy due to the smallest indicators of the mean absolute percentage error (MAPE). The mean absolute error (MAE) and the root mean square error (RMSE) also indicate that the SMOreg algorithm has high prediction accuracy. The results of the analysis showed that machine learning algorithms have achieved highly accurate forecasting efficiency. It was found that nonlinear models cope much better with forecasting export-import operations than linear models. The overall accuracy of the SMOreg algorithm was better for the entire base period interval and the selected forecast period. The results of this analysis can help economic experts in assessing the performance of foreign economic transactions in Ukraine. Implementation of forecasting of export-import operations based on the use of the SMOreg algorithm can be automated to create an expert system to assess the performance of foreign economic transactions in terms of individual regions.