Machine Learning Approaches for Predicting the Elastic Modulus of Basalt Fibers Combined with SHapley Additive exPlanations Analysis

The elastic modulus of basalt fibers is closely associated with their chemical composition. In this study, eight machine learning models were developed to predict the elastic modulus, with hyper-parameter tuning implemented through the GridSearchCV technique. Model performance was evaluated using the coefficient of determination (R2), root-mean-square error (RMSE), and mean absolute error (MAE). SHAP analysis was employed to uncover the relevance of oxide compositions and their interactions with the elastic modulus. Among these models, the Categorical Boosting algorithm exhibited the best results, with an R2 of 0.9554, an RMSE of 4.7556, and an MAE of 2.0323. SHAP analysis indicated that CaO had the most significant influence on elastic modulus predictions. The importance of other oxides was ranked as follows: SiO2, Al2O3, MgO, K2O, Na2O, Fe2O3, FeO, and TiO2. Additionally, SHAP analysis determined oxide ranges for positive elastic modulus prediction. This research provides new insights into leveraging machine learning to optimize the mechanical properties of basalt fibers.