Integrating meteorological and breeding data to predict maize yields using machine learning algorithms

Accurate crop yield prediction is vital for ensuring global food security, particularly amid growing environmental challenges such as climate change. Although deep learning (DL) methods have shown potential in yield prediction, they often demand large datasets and considerable computational resources.

To address these limitations, this study developed a machine learning (ML) model to predict maize hybrid yields by integrating meteorological data with breeder-level genetic information, specifically breeding values estimated using the best linear unbiased prediction (BLUP) method. Four commonly used ML algorithms-Random Forest (RF), XGBoost, Support Vector Regression (SVR), and Gaussian Process Regression (GPR)-were evaluated and optimized through hyperparameter tuning.

Among these models, the RF algorithm achieved the best performance, with a coefficient of determination (R²) of 0.64, a root mean square error (RMSE) of 1010.59 kg/ha, a mean absolute error (MAE) of 743.89 kg/ha, a relative RMSE (RRMSE) of 10.32%, and a mean absolute percentage error (MAPE) of 8.3%.

These results demonstrate that the proposed RF-based model can provide accurate yield predictions for specific maize cultivars under diverse planting conditions. This predictive framework offers practical support for farmers in selecting well-adapted hybrids and serves as a cost-effective, efficient tool for breeders to identify high-yielding maize hybrids optimized for particular environments. Consequently, the model promotes smarter breeding strategies and more precise cultivation recommendations.