Multivariate adaptive regression splines model to predict strength characteristics of high performance concrete

Abstract

Numerous advanced computational tools have emerged for the modelling and analysis of nonlinear multivariate engineering problems. During the recent years, a number of statistical tools are developed to predict the properties of concrete in which artificial neural networks are shown to offer a promising solution for the nonlinear relationship between the variables. This study proposes a Multivariate Adaptive Regression Splines (MARS ) model to predict the compressive strength, split tensile strength and flexural strength of high performance concrete. This model employs cement, silica fume, natural sand and manufactured sand as an input variable and compressive strength, split tensile strength and flexural strength as an output variable. Three individual MARS models have been developed and trained with about 70% of the total 36 data sets and tested with about 30% of the total data sets. The predicted values of compressive strength, split tensile strength and flexural strength are in good agreement with experimental values. It is concluded that MARS model is more reliable and computationally efficient for the prediction of compressive strength, split tensile strength and flexural strength of high performance concrete.