Innovative modeling framework of chloride resistance of recycled aggregate concrete using ensemble-machine-learning methods

Abstract

This study investigates the feasibility of introducing machine learning algorithms to predict the diffusion resistance to chloride penetration of recycled aggregate concrete (RAC). A total of 226 samples collated from published literature were used to train and test the developed machine learning framework, which integrated four standalone models and two ensemble models. The hyperparameters involved were fine-tuned by grid search and 10-fold cross-validation. Results showed that all the models had good performance in predicting the chloride penetration resistance of RAC and among them, the gradient boosting model outperformed the others. The water content was identified as the most critical factor affecting the chloride ion permeability of RAC based on the standardized regression coefficient analysis. The model’s interpretability was greatly improved through a two-way partial dependence analysis. Finally, based on the proposed machine learning models, a performance-based mixture design method and a service life prediction approach for RAC were developed, thereby offering novel and robust design tools for achieving more durable and resilient development goals in procuring sustainable concrete.