Optimization of rice supply chain through multivariate forecasting using ensemble learning models

Abstract

Accurate forecasting of crop yields and production levels is critical for optimizing agricultural supply chains and enhancing food security. This study evaluates six ensemble learning techniques for simultaneously forecasting the yield rate and production of rice utilizing historical data of several features from various districts in Bangladesh. The models are assessed based on five performance metrics. Among the models, Gradient Boosting (GB) demonstrated superior performance, achieving the lowest mean absolute error (MAE), mean squared error (MSE), median absolute error (MeAE), and the highest R2 score of 0.9943. Category Boosting (CatBoost) and Extreme Gradient Boosting (XGBoost) models also performed strongly, with R² values of 0.9917 and 0.9892, respectively. These findings offer significant contributions to enhancing supply chain efficiency by enabling better resource allocation, demand planning, and distribution strategies. Furthermore, the outcomes have extensive implications for policymakers, stakeholders, and food security initiatives, supporting informed decision-making in the face of growing demand and environmental challenges.