Prediction of Compressive Strength of Concrete with Manufactured Sand by Ensemble Classification and Regression Tree Method

Abstract

Manufactured sand (MS) has been increasingly used as fine aggregate for concrete. This paper proposes a prediction of the compressive strength of concrete with manufactured sand (MS-concrete) based on an ensemble classification and regression tree (En_CART) method. A data set containing 1,350 original measured strengths of 328 concrete mixtures from actual engineering projects were used for training and testing. The cross-validation and experimental data from the literature were also used for validation, both indicating that the En_CART model provides an accurate and robust prediction. The comparison of En_CART with various machine learning methods, including artificial neural network, linear regression, Gaussian process regression, random forest, and support vector machine regressions, indicates that the En_CART model indicates superiority in predicting the compressive strength of MS-concrete. Based on the proposed model, the evolution of compressive strength is analyzed. The importance analysis indicates that age is the most significant factor influencing the compressive strength of MS-concrete, and stone powder content presents approximately 25% of the age contribution. The compressive strength of MS-concrete was found to first increase and then decrease with increasing content of MS. The optimal content of MS slightly increases with an increase in the strength level of MS-concrete. Stone powder, at certain MS content, is also found to indicate remarkable improvement in the compressive strength of MS-concrete. The optimum content of stone powder in MS is higher for MS-concrete with lower strength and lower for MS-concrete with higher strength.