Application of machine learning techniques to predict the compressive strength of steel fiber reinforced concrete

Abstract The accurate prediction of compressive strength (CS) in steel fiber reinforced concrete (SFRC) remains a critical challenge due to the material’s inherent complexity and the nonlinear interactions among its constituents. This study presents a robust machine learning framework to predict the...

Full description

Saved in:
Bibliographic Details
Main Authors: Ala’a R. Al-Shamasneh, Arsalan Mahmoodzadeh, Faten Khalid Karim, Taoufik Saidani, Abdulaziz Alghamdi, Jasim Alnahas, Mohammed Sulaiman
Format: Article
Language:English
Published: Nature Portfolio 2025-08-01
Series:Scientific Reports
Subjects:
Online Access:https://doi.org/10.1038/s41598-025-16516-1
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Abstract The accurate prediction of compressive strength (CS) in steel fiber reinforced concrete (SFRC) remains a critical challenge due to the material’s inherent complexity and the nonlinear interactions among its constituents. This study presents a robust machine learning framework to predict the CS of SFRC using a large-scale experimental dataset comprising 600 data points, encompassing key parameters such as fiber characteristics (type, content, length, diameter), water-to-cement (w/c) ratio, aggregate size, curing time, silica fume, and superplasticizer. Six advanced regression-based algorithms, including support vector regression (SVR), Gaussian process regression (GPR), random forest regression (RFR), extreme gradient boosting regression (XGBR), artificial neural networks (ANN), and K-nearest neighbors (KNN), were benchmarked through rigorous model validation processes including hold-out testing, K-fold cross-validation, sensitivity analysis, and external validation with unseen experimental data. Among the tested models, GPR consistently outperformed all others, achieving a maximum coefficient of determination (R²) of 0.93 and the lowest root mean square error (RMSE) of 16.54, thereby demonstrating superior capability in capturing the underlying nonlinear relationships within the data. The generalization performance of each model was examined by systematically altering input variables (fiber type, fiber content, w/c ratio, and aggregate size) while holding other parameters constant. GPR showed remarkable agreement with empirical trends across all validation cases, accurately identifying strength peaks and non-linear behavioral shifts, such as the parabolic relationship between w/c ratio and CS. Models like XGBR, SVR, and RFR provided reasonable estimates but lacked the precision of GPR under complex conditions. In contrast, ANN and KNN demonstrated weaker performance, frequently underpredicting or failing to capture key trends. By leveraging the predictive power and interpretability of advanced machine learning models, this research promotes a paradigm shift in structural engineering workflows.
ISSN:2045-2322