AutoGluon-enabled machine learning models for predicting recycled aggregate concrete's compressive strength

The use of recycled aggregates (RA) is promising in concrete production for construction projects. Therefore, accurately predicting the compressive strength (CS) of this concrete is crucial for understanding the behaviour of such concrete. This study marks an inaugural effort to evaluate the AutoGluon framework for predicting CS for RA concrete. In this novel framework, categorical boosting (CatBoost), light gradient boosting machine (LightGBM), weighted ensemble (WeightENS) and extreme gradient boosting (XGBoost) were considered. The variables used for the modelling were curing days, quantity of cement, water, recycled aggregate, coarse aggregate and fine aggregate. Various statistical and graphical tools were used to evaluate the effectiveness of the proposed models. The findings showed that the CatBoost model achieved the highest prediction performance with a minimal root mean squared error (RMSE) of 1.45. The feature importance evaluation indicated that cement is the most influential variable for CS models. Overall, the study highlights that AutoGluon provides a reliable and robust framework for modelling in structural engineering. Additionally, AutoGluon eliminates the need for an exhaustive and time-intensive process of hyperparameter optimisation.