Crop yield stack: A multi-model ensemble approach for enhanced agriculture forecasting

Reliable crop production forecasts are essential to the agricultural decision-making processes undertaken at local, regional, and global scales. Traditional models (e.g., XGBoost, Elastic Net, Linear Regression) frequently overlook the complexity of yield as influenced by soil, climate, and farm practices. In this paper, we introduced a stacked ensemble model. The model used two meta-learners (i.e., AdaBoost and Gradient Boosting) which were optimized using parametric tuning, and four base-learners (i.e., Linear Regression; Elastic Net; Decision Tree; K-Nearest Neighbours). The performance of the model was evaluated using RMSE, MAE, MSE, and R2 score. Whilst AdaBoost produced an R2 value of 0.94, the better outcomes produced by Gradient Boosting (RMSE = 0.0542; MAE = 0.0346; MSE = 0.0029; R2 = 0.9911) demonstrated the potential of the ensemble. Future work will explore deep learning models to facilitate yield forecasts for various crops in different climates and improve accuracy.