Metaheuristic-optimized machine learning models for accurate prediction of limestone calcined clay cement (LC3) compressive strength

Sanjog Chhetri Sapkota, Ajad Shrestha, Waiching Tang,
Metaheuristic-optimized machine learning models for accurate prediction of limestone calcined clay cement (LC3) compressive strength,
Case Studies in Construction Materials,
Volume 23,
2025,
e05303,
ISSN 2214-5095,
https://doi.org/10.1016/j.cscm.2025.e05303.
(https://www.sciencedirect.com/science/article/pii/S2214509525011015)
Abstract: This research utilizes hybrid machine learning algorithms to accurately predict the compressive strength of Limestone Calcined Clay Cement (LC3). Metaheuristic optimization algorithms, namely Pelican Optimization (PO) and Leopard Seal Optimization (LS), in conjunction with Random Forest (RF) and Extreme Gradient Boosting (XGB) models, to enhance the predictive performance and model interpretability. The RF-PO model achieved a coefficient of regression (R²) of 0.9806 and a root mean square error (RMSE) of 0.0308 MPa during training, while the XGB-LS model outperformed in testing with an R² of 0.9724 and an RMSE of 0.041 MPa, demonstrating high predictive accuracy. Both models exhibited high variance account for (VAF) values exceeding 97 % and low mean absolute error (MAE), indicating their efficiency and accuracy. Feature importance analysis using Shapley Additive Explanations (SHAP) revealed that Age, Coarse Aggregate (CA), and Ordinary Portland Cement (OPC) were the most influential factors, with SHAP values of 11.0, 8.47, and 6.21, respectively. Additionally, this study presents a scalable graphical user interface (GUI) framework for predicting the compressive strength of LC3, supporting practical adoption in industry.
Keywords: Limestone calcined clay cement (LC3); Machine learning; Metaheuristic optimization; Compressive strength prediction; Shapley additive explanations (SHAP)