Machine-Learning-Based Concrete Strength Prediction Considering Effect of High-Temperature Exposure and Supplementary Cementitious Material (SCM) Content

Exposure to high temperature such as during a fire induces physical and chemical changes in concrete, leading to a significant reduction in its compressive strength. Experimental studies for the strength of heated concrete are not only time-consuming but also limited in experimental conditions to account for a wide ranges of parameters, including concrete mix ratios. Even though there have been attempts to predict the compressive strength of concrete exposed to high temperatures using machine learning (ML) models implementing ensemble algorithms, only a few studies have evaluated the prediction performance of different ensemble models. Hence, this study aimed to suggest the most suitable ensemble ML model for predicting the compressive strength of concrete having supplementary cementitious material (SCM) exposed to high temperatures by comparing five ensemble algorithms: gradient boosting regressor (GBR), extreme gradient boosting regressor (XGBR), categorical gradient boosting (CatBoost), random forest (RF), and extra trees (ET). The results revealed that the CatBoost algorithm had the highest predictive accuracy, as measured by the coefficient of determination (R²), root mean squared error (RMSE), and mean absolute error (MAE). A feature importance analysis was also performed to identify the most influential parameters, showing that the water-to-binder ratio and heating temperature were the most significant factors. Finally, the CatBoost and ET algorithms were chosen to predict the strength of heated concrete from new input data that had not been included in the training or testing process. Among them, CatBoost exhibited the better agreement with the experimental results.