Prediction of compressive and flexural strengths of ultra-high-performance concrete (UHPC) using machine learning for various fiber types

Milad Bolbolvand, Seyed Mehdi Tavakkoli, Farshid Jandaghi Alaee,
Prediction of compressive and flexural strengths of ultra-high-performance concrete (UHPC) using machine learning for various fiber types,
Construction and Building Materials,
Volume 493,
2025,
143135,
ISSN 0950-0618,
https://doi.org/10.1016/j.conbuildmat.2025.143135.
(https://www.sciencedirect.com/science/article/pii/S0950061825032866)
Abstract: This study aims to estimate the compressive and flexural strengths of ultra-high-performance concrete (UHPC) with various parameters, such as fibers including steel, basalt, glass, and polypropylene fibers, by using machine learning methods. Several machines learning algorithms, including gradient boosting (GB), light gradient-boosting machine (LightGBM), extreme gradient boosting (XGBoost), categorical gradient boosting (CatBoost), extremely randomized trees (ERT), and deep neural networks (DNN), are employed to predict the strengths. The results are interpreted using shapley additive explanations (SHAP). A total of 321 and 863 experimental data points for flexural and compressive strength, respectively, are considered. Based on the results, the R2 values of 0.9309 and 0.9210 for the test data indicated that CatBoost is the most effective predictor of the model for both compressive and flexural strength. A graphical user interface (GUI) is developed to make the written program more accessible. The study's conclusions are valuable for construction applications and comprehending the elements of UHPC, offering designers and builders practical insights. The source codes and the data files are also available at https://github.com/MiladBolbolvand/UHPC-Machine-Learning.git.
Keywords: Ultra-high-performance concrete; Machine Learning; Compressive strength; Flexural strength; Concrete mix design