Machine learning based boosting models for predicting flexural strength of steel fiber reinforced concrete

Abstract

Steel Fibre Reinforced Concrete (SFRC) is a composite material that exhibits increased toughness, crack resistance and post-cracking behavior as a result of steel fibers. However, compared to regular concrete, the development of strength prediction algorithms for SFRC is still in its infancy because of its complexity and the lack of available data. Flexural strength is an important parameter in the structural durability of SFRC, especially in pavements, tunnel linings and precast structures. The performance of two Machine Learning methods, Extreme Gradient Boosting (XGBoost) and Gradient Boosting Machine (GBM) is investigated in this research work to predict the flexural strength of steel fiber-reinforced concrete. Machine learning has been demonstrated to be a useful tool in civil engineering to simulate the complex, nonlinear behavior of materials such as SFRC. To investigate this capability, a database containing ninety two experimental observations compiled from the literature on the flexural strength of SFRC was used for model training and testing. Gradient Boosting algorithm predicts the flexural strength of SFRC with R² score and RMSE values of 0.992 and 0.242 for training data and 0.941 and 0.851 for testing data respectively. Extreme Gradient Boosting algorithm predicts the flexural strength of SFRC with R² score and RMSE values of 0.993 and 0.239 for training data and 0.933 and 0.902 for testing data respectively. The findings indicated that both GBM and XGBoost had high predictive accuracy.