Prediction of the sulfate resistance for recycled aggregate concrete based on ensemble learning algorithms

Abstract

Recycled aggregate concrete (RAC) has been acknowledged as an effective way to achieve green building and meet the sustainable development goal. Understanding its durability evolution under complicated environments is indispensable to promote its engineering application. This paper investigates the prediction of the sulfate resistance for RAC based on ensemble machine learning approaches. Four ensemble learning methods(random forest, adaptive boosting, gradient boosting, and extreme gradient boosting) were employed to establish the predictive model. 10 variables related to material properties and environmental conditions were selected as inputs. The compressive strength loss is used to quantify the sulfate resistance and also set as the output. A database containing 143 samples was assembled and divided into the training set and testing set. The five-fold cross-validation was introduced for model training and hyperparameters optimization. Results show that all ensemble learning methods can predict the sulfate resistance of RAC with high accuracy and obtain superior performance over standalone machine learning methods, among which the extreme gradient boosting model performs best. Feature importance analysis results indicate that the sulfate resistance of RAC is susceptible to environmental conditions under a dry state among all input variables. The partial dependence analysis of critical parameters verified the robustness of the proposed model. A graphical user interface was further developed to facilitate the use of the machine learning model for the durability design of RAC in the sulfate environment.