Аннотация
Crop yield prediction is important for food security, agricultural planning, and effective resource management, particularly in Northern Kazakhstan, where agricultural production is strongly influenced by climate variability. Changes in temperature and precipitation can affect crop productivity and cause considerable differences in yield between years. The aim of this study is to develop, compare, and evaluate machine learning regression models for wheat yield prediction in the North Kazakhstan Region using climatic data. A dataset covering 1990–2025 was prepared using official crop yield statistics from the Bureau of National Statistics of the Republic of Kazakhstan and climatic data from the NASA POWER database. Daily climatic data were collected from five representative geographical points across the North Kazakhstan Region. The climatic variables included average, minimum, and maximum air temperature and precipitation. Daily temperature values were aggregated into annual mean values, while precipitation was summed to obtain annual totals. The annual values from the five geographical points were then averaged to obtain representative regional indicators. Several regression algorithms were compared, including Linear Regression, Random Forest, Gradient Boosting, Bagging Regressor, Decision Tree, and XGBoost. A hybrid model combining XGBoost and Gradient Boosting was also evaluated. Wheat yield was used as the target variable, while climatic variables were used as predictors. To preserve the temporal structure of the data and avoid information leakage, walk-forward validation was applied for 2020–2025. Model performance was evaluated using R², RMSE, and MAE. XGBoost demonstrated the best predictive performance, achieving R² = 0.811, RMSE = 1.436 c/ha, and MAE = 1.160 c/ha. The results demonstrate the potential of machine learning and climatic data for wheat yield prediction and agricultural decision support in Northern Kazakhstan.