Comparison of Machine Learning Methods with Optimization for Paddy Production Prediction

Abstract

Food security remains a critical challenge in Indonesia, where rice serves as the primary staple and demand continues to rise with population growth. Fluctuations in paddy production pose significant risks to supply stability, highlighting the need for accurate and reliable forecasting models. This study presents a comparative evaluation of Random Forest and Support Vector Regression (SVR) for paddy production prediction using national-level production data aggregated from provincial statistics in Indonesia, incorporating Grid Search and Random Search to optimize model performance. Experimental results demonstrate that SVR optimized with Random Search achieves superior predictive accuracy, yielding an RMSE of 27,478.58 and a MAPE of 0.05%, indicating both low absolute and relative errors. This performance suggests that SVR is more effective in modeling the non-linear and continuous dynamics inherent in paddy production data. Furthermore, Random Search consistently outperforms Grid Search, reflecting its ability to efficiently explore complex hyperparameter spaces. These findings underscore the critical role of both model selection and optimization strategy in improving forecasting reliability. The proposed approach provides a robust framework for data-driven agricultural planning and offers practical value for policymakers in managing food supply, reducing uncertainty, and enhancing national food security.