Predicting self-healing efficiency in recycled aggregate concrete using optimized machine learning models.

Self-healing concrete can repair microcracks and enhance durability, but traditional methods are costly and complex. Using recycled coarse aggregate (RCA) as a substitute for natural aggregates can reduce costs, improve self-healing properties, and support sustainable construction. This study applied machine learning (ML) to predict the impact of RCA on concrete's self-healing performance. A database of 173 datasets was established, with eight key indicators as input variables and the self-healing rate as the output. An optimized NRBO-XGBoost model was developed and compared with four ML models and two optimization methods. The NRBO-XGBoost model outperformed all others, achieving the highest accuracy (R² = 0.9569, RMSE = 7.1800, MAE = 4.9575). Sensitivity analysis using the Shapley method identified crack width as the most influential factor in self-healing performance, while RCA had minimal impact within the studied range. However, RCA's low cost and environmental benefits highlight its potential for practical applications. This study provides theoretical insights into self-healing recycled concrete and introduces an optimized ML approach for performance prediction. The findings offer valuable guidance for sustainable construction practices.