A comparison study of regression analysis for estimating the capillary water absorption of construction stones

Abstract

Large-scale and time-consuming testing is required to establish the rock’s capillary water absorption (${\text{CWA}}$); however, predictive technologies may lessen these needs. In this study, the hybrid support vector regression (${\rm SVR}$) analysis was developed to predict the ${\rm CWA}$ of the building stones. ${\rm SVR}$’s great performance depends on determining the main parameters of it, where for this aim, two optimization algorithms were considered named arithmetic optimization algorithm (${ {\rm AOA}}$) and sine–cosine algorithm (${ {\rm SCA}}$). The mentioned hybrid regression analysis has not been developed for predicting the ${\rm CWA}$ of building rock so far, which makes this study to be innovative. During the train and test rounds, there was a strong correlation between the observed and predicted ${\rm CWA}$ s, with $R^{2}$ values of 0.9668 and 0.9497, respectively. ${\rm AOA} - {\rm SVR}$ may receive less data than ${\rm SCA} - {\rm SVR}$ network by decreasing from 24.478 to 21.759 ${\text{g}}/{\text{m}}^{2} /{\text{s}}^{0.5}$ in the train round and from 22.52 to 18.87 ${\text{g}}/{\text{m}}^{2} /{\text{s}}^{0.5}$ in the test round when using the root mean squared error (${\rm RMSE}$) measure. The performance index ($PI$) shows that the ${\rm AOA} - {\rm SVR}$ system's capacity to represent lower values than ${\rm SCA} - {\rm SVR}$, with train round values of 0.1554 lower than 0.1751 and test round values of 0.1892 lower than 0.227. In comparison with the earlier study, the coefficient of determination ($R^{2}$) values increased from 0.708 (multiple linear regression) to 0.9745 (${\rm AOA} - {\rm SVR}$), demonstrating a considerable improvement in workability. The created ${\rm AOA} - {\rm SVR}$ method may be called the outperformed approach since it can determine the ${\rm CWA}$ of building stones.