Modeling compressive strength and environmental impact points of fly ash-admixed concrete using data-driven approaches

This study examined the capability of white-box machine learning methods in the intelligent design of concrete technology. Therefore, three data-driven methods, multivariate adaptive regression splines (MARS), gene expression programming (GEP), and group method of data handling (GMDH) approaches, were adopted to model the compressive strength (CS) and environmental impact points (P) of fly ash admixture concrete. The main feature of the proposed methods is that they provide formulas for predicting CS and P. The study's findings indicated the acceptable performance of the suggested methods in concrete technology. In general, the MARS approach for the estimation of CS is more acute than the GMDH and GEP approaches. In addition, MARS had results similar to those of the evolutionary polynomial regression (EPR) model generated in the earlier research to predict CS. Moreover, the MARS model performs slightly better than EPR for predicting P. It is noteworthy that MARS presented more straightforward equations than EPR for predicting CS and P. Sensitivity analysis indicated a more effective parameter on CS and P. The accuracy of the developed models was assessed through statistical parameters and scatter, Taylor, and Violin plots. The presented predictive models can have practical applications in the construction of buildings.