Use of regression and machine learning approach in predicting the strength of GBFS‐based geopolymer mortars without experimentation

The determination of the compressive strength (CS) of mortar samples by lengthy experimental tests leads to time and economic losses. Therefore, the predictability of the CS of mortar specimens without conducting tests has gained importance. In this study, the potential for predicting the CS of geopolymer mortars based on granulated blast furnace slag (GBFS) was investigated without the need for experimental work. For this purpose, two model combinations with different independent variables were used. In the first model, molarity, sample age, and curing temperature were set as independent variables, while in the second model, ultrasonic pulse velocity, unit volume weight, porosity, and water absorption properties were included in addition to these variables. In the modeling studies, 108 data sets were used, which were obtained as a result of experimental studies. The prediction of CS values was performed using conventional regression analysis (CRA), multivariate adaptive regression spline (MARS), random forest (RF), multiple additive regression trees (TreeNet), generalized path searcher, and artificial neural network (ANN). The predictive performance of the models was evaluated using the root mean square error (RMSE), the mean absolute error (MAE), and the Nash–Sutcliffe performance statistic (NS). The obtained performance values show that regression‐based methods provide higher values for Model 1, while the highest performance values are obtained with the MARS method. However, when examining the ANN model results, it is clear that the performance values obtained with Model 1 are higher, which means that the CS values can be estimated with high accuracy without the need to conduct experimental studies. According to the results of the ANN10 model, which provided the highest performance values, the smallest RMSE and MAE and the largest NS values for the test dataset were calculated to be 0.464, 0.367 (MPa), and 0.997, respectively. These models offer considerable advantages in terms of both economy and time. In addition, the CS values of geopolymer mortars at different curing temperatures and molarities can be predicted in advance without having to wait until the concrete has been poured and cured for 7–28 days.