A LightGBM-SHAP Framework for Accurate and Interpretable Crop Yield Prediction

Accurate crop yield prediction plays a vital role in agricultural planning and food security, especially in the face of erratic weather and soil conditions. This paper presents an AI-based platform that leverages historical meteorological and soil datasets to forecast crop yield using machine learning techniques. The model integrates key agronomic variables such as rainfall, temperature, nitrogen-phosphorus levels, and crop phenology to train a LightGBM regressor. The system also incorporates explainable AI (XAI) methods to enhance transparency and decision- making utility for farmers. Results show that our approach achieves a coefficient of determination (R2) of 0.94 and an RMSE of 0.72, outperforming traditional statistical models. The system’ s interpretability, facilitated by SHAP values, empowers stakeholders to understand how environmental factors affect yield outcomes. This study demonstrates the potential of data-driven agriculture to address productivity challenges, especially under climate uncertainty.