| description abstract | The accurate prediction of residual compressive strength (RCS) of concrete plays a critical role in assessing concrete constructions’ safety and structural integrity following exposure to elevated temperatures. Existing ensemble models exhibit RCS prediction capabilities, yet they are constrained by their opaque nature. This research endeavors to develop an intelligible model for RCS by employing five ensemble machine-learning models, namely, random forest (RF), adaptive boosting (AdaBoost), gradient boosting (GBoost), light gradient boosting (LGBoost), and extreme gradient boosting (XGBoost), and integrating Shapley additive explanations (SHAP) to ascertain the precise importance of each input variable in forecasting the RCS of concrete under elevated temperature conditions. The input variables encompass concrete type, compressive strength, aggregate type, water-cement ratio, heating type, heating rate, maximum core temperature, and cooling type. Model performance is appraised using established performance metrics such as mean absolute error (MAE), mean squared error (MSE), root-mean squared error (RMSE), and coefficient of determination (R2). The analytical results exhibit the efficacy of employing machine-learning models in accurately predicting the RCS of concrete under elevated temperature conditions. Among the implemented models, XGBoost demonstrated the highest performance, yielding an R2 value of 0.876, closely trailed by the LGBoost model at 0.871. The SHAP analysis elucidates the crucial role of core temperature, water-cement ratio, heating rate, and compressive strength in determining the RCS of concrete. | |
