Predicting the Twenty-Eight Day Compressive Strength of OPC- and PPC-Prepared Concrete through Hybrid GA-XGB Model

This research set out to create a computational tool for predicting the 28-day compressive strength (CS) of concrete prepared from ordinary Portland cement (OPC) and Portland pozzolana cement (PPC) cement. 1,062 datasets of concrete were collected from laboratory experiments. From 1,062 datasets, 524 samples belonged to OPC and 538 samples belonged to PPC. eXtreme gradient boosting (XGBoost) algorithm optimized with genetic algorithm (GA) was utilized for developing an efficient model. The R value obtained for ${\mathrm{GA}\text{-}\mathrm{XGB}}_{\mathrm{OPC}}$ and ${\mathrm{GA}\text{-}\mathrm{XGB}}_{\mathrm{PPC}}$ models in training (TR) and testing (TS) dataset are almost $\ge 0.90$. Mean absolute error (MAE) and root mean square error (RMSE) values obtained for the ${\mathrm{GA}\text{-}\mathrm{XGB}}_{\mathrm{OPC}}$ model were 2.155 and 2.923, respectively. Similarly, the values for the ${\mathrm{GA}\text{-}\mathrm{XGB}}_{\mathrm{PPC}}$ model were 1.815 and 2.888, respectively. The developed models were found to predict more than 90% of the observations within $\pm 20\%$ variations. The level-1 and level-2 validation results certify the ${\mathrm{GA}\text{-}\mathrm{XGB}}_{\mathrm{OPC}}$ and ${\mathrm{GA}\text{-}\mathrm{XGB}}_{\mathrm{PPC}}$ models’ generalizability. Ultimately, a user-friendly laborless and financially feasible computer software or tool was generated in Python for assistance to the field and design engineers in estimating the CS of concrete.