GDP Growth Forecasting Using Machine Learning: Global Evidence with a Kazakhstan Case Study

Zhanar Lamasheva, Valeriy Lakhno, Aizhan Nazyrova, Aibolatkyzy Karlygash, Raiymbek Durbenov

2026

Full text Open access: Hybrid Metadata (CSL-JSON)
DOI 10.22266/ijies2026.0630.33
Data confidence 70 medium
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

Forecasting GDP growth remains a difficult task in applied macroeconomics, particularly in commoditydependent economies that are highly exposed to external shocks.This study develops an empirical forecasting framework that combines supervised machine learning, unsupervised clustering, and classical time-series methods.The analysis draws on a global panel of 196 countries spanning 1980-2024 and a feature-engineered dataset containing 7,620 observations.In the supervised learning stage, Random Forest, Gradient Boosting, and SVM with an RBF kernel are evaluated for recession classification, while several regression models are employed to predict continuous GDP growth.For the country-specific forecasting exercise, ARIMA, Prophet-style, and LSTM-style specifications are applied to Kazakhstan.To preserve strict temporal ordering and reduce the risk of information leakage, model evaluation relies on a chronological train-test split, with 2015 used as the cut-off year.The results show that Gradient Boosting delivers the strongest overall performance.In the classification task, it achieves an accuracy of 97.9% and a ROC-AUC of 0.997.In the regression task, it records an RMSE of 2.274 and an R² of 0.852.Feature-importance analysis indicates that rolling-average variables over three-and five-year horizons account for the largest share of predictive information, highlighting the importance of medium-term growth dynamics.In the Kazakhstan case study, the ARIMA(3,1,0) model projects annual GDP growth in the range of 4.3% to 5.5% through 2029, broadly in line with IMF benchmark projections over the near term.The study contributes to the literature by integrating classification, regression, clustering, and country-level forecasting within a single empirical framework.It also shows that parsimonious machine learning models based largely on GDP-derived features can provide useful signals for recession detection and growth forecasting in cross-country settings.From a practical perspective, the results are relevant for early-warning and monitoring systems in data-constrained, resource-dependent economies.