Statistical Methods and Machine Learning Approaches for Unemployment Rate Forecasting: A Comparative Analysis

Parshakova Daria, Yin Gan

Abstract: The study presents a comparative analysis of econometric and machine learning methods for forecasting unemployment rates across Russian regions. Traditional models including linear regression, ARIMA, and VAR are compared with decision trees, random forests, gradient boosting methods, and neural network architectures. Regional data covering 2000 to 2024 are employed with extensive preprocessing, feature selection, and walk forward validation. The results indicate that LightGBM achieves the best performance for annual forecasting with R² = 0.785 and a 46.4% RMSE reduction relative to the linear baseline, while LSTM performs competitively for monthly forecasting. The findings demonstrate the importance of feature selection, accounting for vintage data, and incorporating regional heterogeneity in unemployment forecasting.

Keywords: unemployment forecasting, machine learning, gradient boosting, LightGBM, LSTM, regional economics, labour market, Russian regions

How to Cite?: Parshakova Daria, Yin Gan, "Statistical Methods and Machine Learning Approaches for Unemployment Rate Forecasting: A Comparative Analysis", Volume 15 Issue 6, June 2026, International Journal of Science and Research (IJSR), Pages: 750-753, https://www.ijsr.net/getabstract.php?paperid=MR26612175850, DOI: https://dx.doi.org/10.21275/MR26612175850

Download Citation: APA | MLA | BibTeX | EndNote | RefMan