Optimizing rice yield prediction: A policy-oriented approach to agricultural sustainability

Accurate forecasting of rice yield is critical to ensuring global food security, supporting market stability, and enabling data-driven agricultural policy. This study evaluates a comprehensive suite of machine learning (ML) and statistical regression models to predict rice yield using meteorological and agronomic data from Kerala, India. Models assessed include Random Forest Regression (RFR), Gradient Boosting Regression (GBR), Support Vector Regression (SVR), Multiple Linear Regression (MLR), Least Absolute Shrinkage and Selection Operator (LASSO) Regression, Ridge Regression (RR), Elastic Net Regression (ELNR), and a hybrid ensemble combining LASSO, XGBoost, and RFR. GBR achieved the highest predictive accuracy ( R 2 = 0.839), followed closely by SVR ( R 2 = 0.837) and the hybrid model ( R 2 = 0.827), with the hybrid model attaining the lowest Root Mean Square Error (RMSE). Beyond predictive performance, the study integrates causal inference to assess the impact of a policy intervention initiated in 2010. Regression results indicate that the policy is associated with a statistically significant yield increase of approximately 279 Kg/Ha ( p < 0.001 ) , while excessive rainfall is negatively associated with yield (–0.243 Kg/Ha/mm; p < 0.001 ). The integration of predictive analytics and policy-aware modeling presents a robust framework for forecasting yield and evaluating the real-world effectiveness of agricultural interventions. These insights offer substantial value for policymakers and stakeholders aiming to optimize food production systems. • Integrated predictive and causal models to assess rice yield and policy impact. • GBR model achieved highest accuracy ( R 2 = 0.839 ) on Kerala rice data. • 2010 policy intervention increased rice yield by ∼279 Kg/Ha ( p < 0.001 ). • Excess rainfall reduced yield by − 0.243 Kg/Ha/mm ( p < 0.001 ). • Framework supports data-driven decisions for food security and crop planning.