Investigation of Explainable Crop Yield Prediction: Leveraging Ensemble Learning and a Novel Blend Model Approach

Abstract

Background Accurate Crop Yield Prediction (CYP) is pivotal for ensuring food security and optimizing agricultural practices. In the face of climate change and resource limitations, precise yield forecasts can help farmers make informed decisions, enhance sustainability, and effectively allocate resources. Methods This study affirms the superior efficacy of Ensemble Learning (EL) models in enhancing CYP accuracy and proposes a novel Blend Model that synergizes predictions from individual base learners (Random Forest, XGBoost, AdaBoost) with established ensemble techniques (Model Averaging, Stacking, Voting Regressor). Results Utilizing a comprehensive dataset encompassing temperature, rainfall, and pesticide usage, this approach is evaluated against established metrics such as Mean Squared Error (MSE), Mean Absolute Error (MAE), R-squared (R²), and Explained Variance. The Blend Model, designed to combine the strengths of base models, achieved an exceptional R² of 0.9899, capturing nearly 99% of the variance in crop yields with minimal errors (MSE: 72,974,685.72, MAE: 3,274.39). While AdaBoost and Stacking models demonstrated effectiveness, the Blend Model outperformed them in precision. Gradient Boosting (R²: 0.8784) and Meta-AdaBoost (R²: 0.9861) showed promise but exhibited higher errors. Conclusion This study, for the first time, investigates Explainable Artificial Intelligence (XAI) techniques—SHapley Additive exPlanations (SHAP), Local Interpretable Model-Agnostic Explanations (LIME), and Explain Like I'm 5 (ELI5)—with EL models to elucidate the critical factors influencing CYP. This research highlights the transformative potential of EL models in agricultural practices, significantly enhancing sustainability and food security. By providing detailed insights into the factors influencing CYP, this study empowers informed decision-making by farmers and policymakers, setting a new benchmark for future research in crop yield prediction.