Performance measurement of experimentally validated machine learning algorithms to predict compressive strength of sustainable concrete using GGBFS and rice husk ash

This study explores the ability of advanced machine learning (ML) models to anticipate the compressive strength (CS) of concrete incorporating GGBFS and rice husk ash (RHA) as cement replacements. A dataset of 359 records with eight input parameters was used to create predictive models such as Adaptive Boosting (ADB), Decision Tree (DT), Gradient Boosting (GB), k-nearest Neighbors (KNN), Light Gradient Boosting (LGB), and Extreme Gradient Boosting (XGB). For experimental assessment, specimens were created by substituting cement with 10% RHA and 10%, 20%, and 30% GGBFS. The experimental outcomes were further used to validate graphical user interface developed in this study. Additionally, microstructural analysis was carried out to highlight the influence of GGBFS and RHA on the concrete’s microstructure. Test results have shown that the mix having 10% RHA and 20% GGBFS combination achieved the highest CS at 7, 28, and 56 days. Model performance was assessed using evaluation measures including RMSE, R 2 , MAPE, and MAE. The XGB model was highly accurate (R 2 of 0.99 and 0.88 for the train and test phase), followed by LGB (R 2 of 0.984 and 0.897 for the train and test phase). Further, SHAP and partial dependence plot analysis confirmed that curing ages, cement, and water content were the most important factors governing CS prediction. Microstructure assessment confirmed that incorporating GGBFS showed lower porosity and more structured C-S-H bonds. By incorporating ML approaches, the construction industry may greatly improve productivity, accuracy, and sustainability, paving the door for smarter, more environmentally friendly practices.