Prediction of mechanical strengths of geopolymer concrete using optimized machine learning models with graphical user interface
Geopolymer concrete (GPC) has emerged as an environmentally friendly alternative to ordinary Portland cement (OPC)... See more
Geopolymer concrete (GPC) has emerged as an environmentally friendly alternative to ordinary Portland cement (OPC)
concrete by utilizing industrial byproducts, thereby reducing CO₂ emissions and promoting sustainable construction. In
this research, machine learning (ML) models such as CatBoost, Extreme Gradient Boosting (XGBoost), Light Gradient
Boosting (LGB), and Extra Trees models were used to predict compressive and flexural strength of geopolymer concrete
(GPC). In this study, a total of ten input parameters were used to predict the compressive and flexural strength of GPC.
The study employed statistical evaluation metrics, including R², RMSE, MAE, and MAPE, to assess model accuracy. ML
models exhibits R² value more than 0.950 at the three stages (train, test and validation) of the prediction of FS and R²
value shows above 0.900 at the three phase (train, test and validation) of the prediction of CS. For CS, ML models had
RMSE values less than 3.1413, 4.7050, and 4.5282 during the training, testing, and validation stages, respectively. The
RMSE value found for ML models during train, test and validation below 0.2392, 0.4160 and 0.3843, respectively for
FS of GPC. SHAP analysis further identified Age, coarse aggregate, and GGBFS as the most influential parameters for
CS, while Fine Aggregate, CA, and Age had the greatest impact on FS. These findings highlight the effectiveness of ML
techniques in predicting GPC performance and emphasize the potential of data-driven approaches in optimizing sustainable
concrete formulations. Additionally, the combined use of SHAP analysis and a user-friendly GUI supports interpretability
and real-time decision-making, contributing to the advancement of eco-friendly concrete technology through robust
predictive modeling and data-driven optimization.
2025-10
