Performance-Based Prediction Of Shear And Flexural Strengths Of Fiber-Reinforced Concrete Beams Using Machine Learning

Abstract

Fiber-reinforced concrete (FRC) is an advanced construction material that enhances the tensile strength, ductility, and cracking resistance of conventional concrete. In this study, a database of 847 experimental tests compiled from the literature published between 1990 and 2025 was used to develop a machine-learning (ML) framework for predicting the shear and flexural strengths of FRC beams. The database spans various fiber types, fiber volume fractions (0.5–3.0%), concrete compressive strengths (20–120 MPa), and beam sizes and shapes. Six ML algorithms  Artificial Neural Networks (ANN), Support Vector Regression (SVR), Random Forest (RF), Gradient Boosting Machines (GBM), Extreme Gradient Boosting (XGBoost), and an Adaptive Neuro-Fuzzy Inference System (ANFIS) were developed and evaluated using key material, geometric, and reinforcement parameters. Model performance was assessed using R², RMSE, MAE, and MAPE. XGBoost showed the highest predictive accuracy among all models, with R² = 0.94 for shear strength and R² = 0.96 for flexural strength, outperforming empirical and code-based prediction methods by up to 50% in mean absolute percentage error. Feature-importance analysis identified concrete compressive strength, fiber volume fraction, and reinforcement ratio as the dominant variables governing beam behaviour. The results demonstrate the capability of ML approaches to capture the complex nonlinear relationships governing FRC beam behaviour and indicate that the developed models can support strength prediction and design optimization with a high degree of reliability, contributing to safer, more economical, and more sustainable construction practice.