Explainable prediction framework for high‐temperature residual strength of SiC _f /SiC composites driven by XGBoost + SHAP

Abstract Aiming at the problem of predicting the residual bending strength of ceramic‒matrix composites at high temperature, this paper proposes an interpretable prediction framework combining extreme gradient boosting (XGBoost) + SHAPley Additive EXPlanations (SHAP). The flexural strength before heat treatment (Bend‐b) of the SiC f /SiC composite material studied in this paper is mainly affected by the modulus and volume fraction (F‐mol and F‐vol) of the reinforcing fiber (SiC f ) during the preparation process, as well as the density (density) and porosity (porosity) after composite SiC matrix. On this basis, the flexural strength after heat treatment (Bend‐a) is affected by the heat treatment parameters (temperature and time) and exhibits varying degrees of performance degradation. After filling missing values, removing outliers, and visualizing feature distribution and correlation, six prediction models, linear, ElasticNet, support vector regression, multilayer perceptron, random forest, and XGBoost, are trained. After optimizing hyperparameters through grid search and fivefold cross‐validation, the model performance is evaluated using R 2 , mean absolute error, and root‐mean‐square error. XGBoost shows the highest accuracy and the lowest error. Subsequently, global and local SHAP analysis is used to clarify the contribution path and size of each feature, revealing the dominant role of F‐vol and porosity in determining Bend‐b, and the joint driving effect of Bend‐b, temperature and time on Bend‐a. The results show that the XGBoost + SHAP framework not only significantly improves the prediction accuracy but also has good interpretability, providing a reliable tool for the design and optimization of the mechanical properties of ceramic‒matrix composites in high‐temperature environments.