Advanced machine learning models for the prediction of ceramic tiles' properties during the firing stage.

The firing stage is a critical phase in ceramic tile production, where the interplay of raw material composition and thermal treatment determine essential properties such as water absorption (WA) and bending strength (BS). This study employs advanced machine learning (ML) models to accurately predict these properties by capturing their complex nonlinear relationships. A robust dataset of 312 ceramic samples was analyzed, including variables such as particle size distribution, chemical and mineralogical composition, and firing temperatures ranging from 1000 to 1300 °C. Among the four ensemble ML models evaluated, CatBoost demonstrated the highest predictive performance. Model accuracy was assessed using multiple evaluation metrics, including the coefficient of determination (R²), root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). To enhance interpretability, SHapley Additive exPlanations (SHAP) were used, revealing that clay mineral content and SiO₂ concentration were the most influential factors for WA, contributing approximately 40% and 30%, respectively. For BS, firing temperature (35%) and Al₂O₃ content (25%) were identified as the key predictors. Partial dependence plots further illustrated critical thresholds, such as a significant drop in WA above 62% SiO₂ and optimal BS values near 1200 °C, dindings that align with known ceramic processing principles while offering new, data-driven formulation insights. These results demonstrate the value of explainable artificial intelligence (AI) in enabling real-time process optimization, enhancing product consistency, and supporting energy-efficient ceramic manufacturing. Future work will focus on extendingthe dataset to include a wider variety of clay compositions and investigating hybrid modeling approaches for further improve predictive performance.