Bandgap Prediction of Silicon Oxide Materials for Electric Furnace Refractories Based on Explainable Machine Learning

An interpretable machine learning framework was constructed to predict the bandgap of silicon oxide materials used for electric furnace refractories. Among seven machine learning models compared, the AdaBoost ensemble model performed the best with an R2 of 0.80 and MAE of 0.5, indicating high accuracy. Breaking through the traditional limitation of directly correlating features such as the conduction-band minimum (CBM) and valence-band maximum (VBM), this study utilized SHapley Additive exPlanations (SHAP) analysis to uncover deeper relationships between critical features like ‘energy above hull’, ‘num of unique magnetic sites’, and ‘formation energy per atom ‘ with the bandgap. These features significantly influence the thermal stability and erosion resistance of the material. This research provides a theoretical basis for optimizing the performance of silicon oxide materials and green metallurgical processes, thereby promoting sustainable development in the steel industry.