Machine learning-based modeling and prediction of methanol production via integrated co-gasification of polymer waste and rice husk

Abstract

Co-gasification of rice husk and polymer waste presents a sustainable pathway for methanol production, addressing waste management challenges while supporting renewable energy demand. This study applies ensemble machine learning models like Extreme Gradient Boosting (XGBoost), Gradient Boosting Regressor (GBR), and Random Forest (RF) to predict key performance indicators, including stoichiometric number and methanol yield. A comprehensive hyperparameter tuning strategy was implemented to enhance predictive accuracy. Among the models, XGBoost demonstrated superior performance, achieving the lowest test mean squared error (MSE) values of 0.0007 for stoichiometric number and 0.0002 for methanol yield, along with high coefficients of determination (R²) of 0.9984 and 0.9889, indicating strong generalization capability. GBR showed moderate performance, while RF exhibited comparatively lower accuracy. Linear Regression performed the worst, with higher MSE values (0.0366 and 0.0017) and lower R² scores. Taylor diagram analysis further confirmed the robustness of XGBoost in matching observed data. The results highlight the effectiveness of ensemble learning in optimizing methanol production and supporting sustainable energy development.

KEY WORDS: Sustainability, Co-gasification, Agricultural waste, Plastic, Waste-to-energy, Machine learning, Ensemble methods

Bull. Chem. Soc. Ethiop. 2026, 40(11), 2287-2304                                            

DOI:  https://dx.doi.org/10.4314/bcse.v40i11.1