ABSTRACT The dissolution behavior of nonmetallic inclusions in slag is critical for determining steel cleanliness. In this study, a machine learning (ML) model was developed to predict inclusion dissolution time. A key contribution of this work is the construction of a comprehensive database by integrating literature‐based experimental data with a diffusion‐distance‐controlled dissolution (DDD) physical model. The performances of five regression algorithms were compared, and the results indicate that the extreme gradient boosting (XGBoost) model provides the best accuracy and stability. For the optimal model, the coefficients of determination ( R 2 ) reached 0.994 for the test set and 0.996 for an independent validation set, with corresponding mean absolute errors (MAEs) of 25.999 and 52.250 s, respectively. This model accurately predicts dissolution times for seven inclusion types (Al 2 O 3 , MgAl 2 O 4 , MgO, TiO 2 , SiO 2 , Al 2 TiO 5 , and CaO·2Al 2 O 3 ). To improve industrial applicability, a reduced feature (“less feature”) strategy was further introduced while maintaining robust performance. Shapley additive explanations (SHAP) analysis indicates that CaO content is the dominant factor (26.51%–50.02%), followed by dissolution degree ( r / r 0 ) and temperature, while other features each contribute less than 10%. The model’s decision logic reliably captures the dominant mechanisms governing inclusion dissolution.
Song et al. (Sun,) studied this question.