To predict the compressive strength of self compacting concrete with recycled aggregates utilizing ensemble machine learning models

Abstract

This study aims to apply machine learning methods to predict the compression strength of self-compacting recycled aggregate concrete. To obtain this goal, the ensemble methods: Random Forest (RF), K-Nearest Neighbor (KNN), Extremely Randomized Trees (ERT), Extreme Gradient Boosting (XGB), Gradient Boosting (GB), Light Gradient Boosting Machine (LGBM), Category Boosting (CB) and the generalized additive models: Inverse Gaussian (GAM1) and Poisson (GAM2) were applied. For the development of the models, 515 research article samples were collected and divided into three subsets: training (360), validation (77), and testing (78). The SCC components: cement, water, mineral admixture, fine aggregates, coarse aggregates, and superplasticizers were taken as input variables and compression strength as output variables. To determine the ability of the models to project compressive strength, the following metrics were used: R2, RMSE, MAE, and MAPE. The results indicate that the RF (R2 = 0.7128, RMSE = 0.0807, MAE = 0.06) and GB (R2 = 0.6948, RMSE = 0.0832, MAE = 0.0569) models have a strong potential to predict the compressive strength of SCC with recycled aggregates. The sensitivity analysis of the RF model indicates that cement and water are the variables that have the highest impact in predicting the compressive strength, while coarse aggregate has the lowest impact.