Predictive modelling of concrete compressive strength incorporating GGBS and alkali using a machine-learning approach

This study presents a comparative analysis of three machine-learning models, namely Random Forest, Gradient Boost Regressor, and AdaBoost Regressor, for predicting the compressive strength of concrete. The Ground-Granulated Blast Furnace Slag (GGBS) fraction and alkali concentration are used as input characteristics in the dataset that is used to train and test the models. The result shows that the Random Forest model trained to the highest coefficient of determination (R-squared) of 0.9636, closely followed by the Gradient Boost Regressor at 0.9631, while the AdaBoost Regressor's R-squared score was slightly lower at 0.9029. The Random Forest and Gradient Boost Regressor models maintained their strong predictive performance for the testing phase, with R-squared values of 0.9411 and 0.9405, respectively. The AdaBoost Regressor showed a comparatively lower R-squared value of 0.86. Additionally, the mean square values for the models for Random Forest, Gradient Boost Regressor, and AdaBoost Regressor, respectively, were 2.9610, 2.9176, and 7.6290. As an indicator of how precisely the predictions have been determined, lower mean square values reflect stronger model performance. The results of the Sobol method's sensitivity study showed that the GGBS percentage had a significant degree of sensitivity in predicting compressive strength. This result highlights the major impact of GGBS on the overall strength properties of concrete. This work demonstrates that machine-learning algorithms can accurately estimate compressive strength using GGBS and alkali concentration as inputs, contributing in the formulation of durable concrete and optimum mixture proportions. © 2023, The Author(s), under exclusive licence to Springer Nature Switzerland AG.