Prediction of the strength characteristics of basalt fibre reinforced concrete using explainable machine learning models

• Published in: Discover applied sciences Vol. 7; no. 8; pp. 1 - 24
• Main Authors: Wickramasuriya, B. Y., Alahakoon, Yasitha, Krishantha, B. R. G. A., Alawatugoda, Janaka, Ekanayake, I. U.
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 05.08.2025; Springer
• Subjects: Applied and Technical Physics; Basalt fibre reinforcement concrete; Chemistry/Food Science; Earth Sciences; Engineering; Environment; Explainable artificial intelligence; Machine learning; Materials Science; Mechanical properties; Random forest; Mechanical properties; Explainable artificial intelligence; Random forest; Basalt fibre reinforcement concrete; Machine learning
• ISSN: 3004-9261; 3004-9261
• DOI: 10.1007/s42452-025-07528-7

Adding basalt fibre to concrete has become a promising way to improve its strength under different loading conditions. Predicting the strength of basalt fibre reinforced concrete (BFRC) is more complex than for conventional concrete. Traditional methods rely on iterative experiments to understand the underlying behaviour. This study applies machine learning (ML) models, paired with explainable artificial intelligence (XAI), to predict the compressive, flexural, and tensile strengths of BFRC efficiently and transparently. Three datasets, each with 267 samples, were used for model development. Each sample included 10 features: cement, fly ash, silica ash, coarse aggregate, fine aggregate, water, superplasticiser, fibre diameter, fibre length, and fibre content. Three ML models, random forest (RF), support vector machine (SVM), and decision tree (DT), were evaluated. RF achieved the highest accuracy for compressive (R 2  = 0.926, MSE = 9.4, MAE = 2.3) and flexural strength (R 2  = 0.882, MSE = 0.5, MAE = 0.5), while DT performed best for tensile strength (R 2  = 0.935, MSE = 0.2, MAE = 0.3). Shapley additive explanation (SHAP)-based explainable artificial intelligence (XAI) methods revealed key predictors: fine aggregate for compressive strength, fibre diameter for tensile strength, and silica ash for flexural strength. Also, the authors developed a web-based online application to predict the strength (compressive, flexural and tensile) to improve the accessibility and usability of the findings. This approach offers a practical alternative to experimental testing and contributes to a better understanding of BFRC behaviour. It holds strong potential for adoption in the construction industry.