Machine learning predicting sintering temperature for ceramsite production from multiple solid wastes

Abstract

An efficient machine learning model was developed to accurately predict the sintering temperature of ceramsite synthesized from various solid waste materials. Based on experimental data from 236 samples, eight key chemical components were defined as input features, and six machine learning models were trained and evaluated. Among them, XGBoost achieved the highest performance, with an R² of 0.950 and an RMSE of 7.767 on the test set, effectively capturing the quantitative relationship between chemical composition and sintering temperature. SHAP analysis revealed that SiO₂ and Al₂O₃ significantly elevate sintering temperature, whereas alkaline oxides such as CaO and MgO contribute to its reduction. Applicability domain analysis showed that all samples had leverage values below the warning threshold and normally distributed residuals, indicating strong generalizability and predictive reliability on unseen data. Beyond delivering a robust predictive framework, the study also offers new insights into the roles of chemical constituents in the sintering process, underscoring the potential of machine learning in optimizing ceramsite production.