Artificial intelligence-based ensemble models with GUI for predicting the compressive strength of waste glass concrete

The sustainable utilization of post-consumer waste glass in concrete has emerged as a promising approach to reduce cement consumption, mitigate landfill disposal, and enhance material performance. However, most previous predictive studies have relied on limited datasets, excluded chemical composition effects, or used single machine-learning algorithms, leading to restricted generalization. This study evaluates and develops artificial intelligence-based ensemble learning models to predict the compressive strength of concrete incorporating waste glass powder (WGP) as a partial cement replacement. A dataset of 337 experimental results was compiled from the literature published between 2007 and 2024, including eleven key input variables such as WGP size and replacement level, water-to-cement ratio (W/C), aggregate properties, curing age, and chemical composition of WGP (SiO 2 , CaO, Na 2 O). Five advanced ensemble algorithms — Gradient Boosting Regressor, Extreme Gradient Boosting Regressor, LightGBM Regressor, CatBoost Regressor, and Histogram-based Gradient Boosting regressor — were trained and optimized using Bayesian hyperparameter tuning and validated with 10-fold cross-validation. Performance was assessed using R 2 , RMSE, MSE, MAE, and MAPE metrics. All models demonstrated excellent predictive ability (R 2 > 0.94), with CatBoost achieving the highest testing accuracy (R 2 = 0.96, RMSE = 2.34 MPa, MAE = 1.63 MPa). Feature importance and SHAP analysis revealed curing time and W/C as the most influential parameters, followed by aggregate content and WGP replacement level. Parametric studies confirmed the expected concrete behavior, with strength gains over curing time and reductions at high WGP replacement and W/C. A graphical user interface (GUI) was developed using the CatBoost model, enabling the practical prediction of compressive strength for various mix designs. The integration of chemical composition–based modeling, ensemble learning optimization, and GUI deployment establishes a practically oriented framework that advances sustainable concrete design and facilitates its broader adoption within the construction industry. • Waste glass powder (WGP) was used as partial cement replacement in concrete. • 338 datasets with 11 input parameters were modeled using ensemble ML algorithms. • CatBoost achieved highest accuracy (R 2 = 0.96, RMSE = 2.34 MPa, MAE = 1.63 MPa). • Curing time and W/C ratio were identified as the most influential factors. • A GUI tool was developed for practical prediction of compressive strength.