LAUSR

Abstract In this paper, in order to predict the flexural strength and compressive strength of cement mortar containing nano-silica (NS) and micro-silica (MS), the possibility of using the support vector machine (SVM) approach was investigated through four different kernels, including radial basis function (RBF), polynomial, linear, and sigmoid. The input parameters were employed based on a dataset containing 32 mixtures, 32 flexural specimens, 480 compressive specimens, and 7 mix design variables, namely water/cement ratio (W/C), sand/cement ratio (S/C), nano-silica/cement ratio (NS/C), micro-silica/cement ratio (MS/C), age and porosity of specimens. Numerical results showed that the RBF kernel generally performs better and gives more accurate results compared to the polynomial, linear, and sigmoid kernels. In case the porosity is considered as an input parameter, the values of the root mean square error of SVM with RBF kernel in the prediction of flexural strength and compressive strength are 0.2909 (correlation coefficient of R2 = 0.9970) and 1.2969 (R2 = 0.9987), respectively. Moreover, the cement mortar strength was predicted using the multilayer perceptron (MLP) neural network, radial basis function (RBF) network, and general regression neural network (GRNN) methods. The obtained results were then compared with the results of SVM based on RBF kernel (SVM-RBF). The comparative evaluation in the SVM models was carried out in two cases: in the first case, the porosity is not considered as an input parameter while in the second case the porosity included in the input parameters. The results indicate when the porosity is considered as the input parameter, higher accuracy and better results would be obtained. The sensitivity analysis was also carried out to evaluate the effects of input parameters against the predicted responses. To validate the proposed SVM models, 86 flexural data as well as 266 compressive data were considered from literature and then the flexural strength and compressive strength of the cement mortar are predicted using the support vector regression (SVR) kernels. The results of this study show that the SVM-RBF is a relatively new, powerful, and alternative method for predicting the flexural strength and compressive strength of the cement mortar containing nano- and micro-silica.