Bài báo khoa học

Compressive strength is a primary factor of concrete. Concrete characteristic severely affects by temperatures, which can reduce the strength properties of the concrete. Therefore, the accurate prediction of the concrete compressive strength at elevated temperatures is challenging. This study aims to develop a robust hybrid SCA-XGB model that integrated a sine cosine algorithm and an extreme gradient boosting model, to precisely predict the compressive strength of concrete at elevated temperatures. First, the database of concrete strength at different temperatures is collected from the literature. Then, hybrid SCA-XGB models are developed with the assistance of the SCA algorithm for fining-tune the hyperparameters of the XGB model for predicting the compressive strength at elevated temperatures. As a result, several hybrid SCA-XGB models are generated by changing the training-test ratio of database and the population size of the SCA algorithm. The best hybrid SCA-XGB model is chosen by evaluating the statistical metrics. The performance of the best SCA -XGB model is compared with those of other machine learning (ML) models. The SCA-XGB model achieves credible results with (0.995, 0.982) of R2, (0.925 and 0.810) of A10, (1.774 MPa and 3.676 MPa) of RMSE, and (1.317 MPa and 2.706 MPa) of MAE. It is found that the SCA-XGB model not only accurately predicts the compressive strength of concrete at elevated temperatures but also outperforms other models. Notably, the black box behind the SCA-XGB model is explored using the SHapley Additive exPlanation (SHAP) method via the global and local explanations. Finally, a web application is built based on the SCA-XGB model for users can rapidly predict the compressive strength of concrete at elevated temperatures.