Efficient creep prediction of recycled aggregate concrete via machine learning algorithms

This paper comprehensively investigated the surrogate model of recycled aggregate concrete (RAC) creep behavior prediction for the first time utilizing five typical machine learning (ML) algorithms trained with the RAC_Creep_v1 database. The grid search algorithm and k-fold cross-validation are performed to find the optimal hyperparameters. Then, attribute importance analysis and correlation analysis were conducted to evaluate the effect of various input variables on the results. By retraining the XGBoost model after feature selection, the results revealed that loading age has the greatest impact on RAC creep performance. The XGBoost, the optimal model via evaluation and comparison, was shown to have a higher efficiency and accuracy for predicting RAC creep subjected to different variables. Furthermore, the result of this paper can assist designers in comprehending the performance of RAC structures, promoting the application of RAC in buildings and the study for reducing carbon emissions.