Bulletin of
Vitebsk State
Technological
University

Abstract

In the context of the digital transformation of the economy and accelerated technological development, traditional approaches to labor market information support show significant limitations, leading to structural imbalances and ineffective employment policies. The relevance of the study stems from the need to improve the methodology for forecasting labor demand in the Republic of Belarus. The aim of the research is to substantiate ways to improve labor market information support based on a comparative analysis of data sources and forecasting methods. To achieve this goal, an analysis of two data sources was conducted: official statistics from the State Employment Service (SES) and data from online job portals (OJP). Forecasting models were built and compared based on classical time series analysis methods (SARIMA) and modern machine learning algorithms (XGBoost, LSTM, Prophet), as well as their hybrid combinations. The study found that the quality and stability of the source data have a decisive impact on forecast accuracy. The research was conducted in three stages: baseline forecasting using SES data (SARIMA, MAPE 11.20 %), application of machine learning models to OJP data (MAPE 22–27 %), and extended modeling using SES data. The best result was demonstrated by the hybrid SARIMAX+XGBoost model using SES data (MAPE 3.15 %), which is 3.6 times more accurate than the baseline model and 7.1 times more accurate than the similar model using OJP data. The study reveals that SES and OJP data have distinct characteristics and application areas: SES data ensure high accuracy of quantitative forecasts due to time series stability, while OJP data provide value for real-time monitoring and qualitative analysis of skills demand structure. Recommendations are proposed for creating a comprehensive labor market information and analysis system integrating both data sources.

Improving the information support of the labor market of the Republic of Belarus based on modern methods of labor demand forecasting