Efficient estimating compressive strength of ultra-high performance concrete using XGBoost model

Abstract

An ensemble technique using XGBoost model is employed to predict the compressive strength of ultra-high performance concrete (UHPC). A 931 UHPC mixture collection with 17 input variables is employed, in which 230 were from laboratory experiments and the rest from scientific literature. The best results are obtained by tuning the hyper-parameters using Pareto multi-objective optimisation to find the optimum values of R2 for both train and test dataset. The obtained solution with R2 = 0.8922, RMSE = 7.860 MPa, MAE = 5.930 MPa show better performance with those from previous study. Partial dependence plots are illustrated to investigate the effects of some important input variables on the compressive strength of UHPC. Graphical User Interface (GUI) is written in Python and provided freely for users to support the design and interpretation the results of UHPC. This model could be benefit in the development of new dosages of UHPC by reducing the time and cost of the experimental campaign, which allows the preselection of components with better response in the model.