Shear capacity prediction of FRP-RC beams using single and ensenble ExPlainable Machine learning models

Abstract

Corrosion in steel reinforcement is a central issue behind the severe deterioration of existing reinforced concrete (RC) structures. Nowadays, fiber-reinforced polymer (FRP) is increasingly being used as a viable alternative to conventional steel reinforcement due to its anti-corrosive nature. The accurate estimation of the shear capacity of FRP reinforced concrete (FRP-RC) elements is critical for a reliable and accurate design and performance assessment of such members. However, existing shear models are often developed based on a limited database and important factors, limiting their prediction effectiveness. Hence, this paper presents novel machine learning (ML) based models for predicting the shear capacity of FRP-RC beams. A total of eleven ML models starting from the simplest white-box models to advanced black-box models are developed based on a large database of FRP-RC beams. Such investigation helps in examining the necessity of complex ML models and identify the most accurate predictive model for the shear capacity of FRP-RC beam. Moreover, a unified framework known as SHapley Additive exPlanation (SHAP) is used to identify the most important factors that influence the shear capacity prediction of FRP-RC beams. Among all investigated ML models, the extreme gradient boosting (xgBoost) model showed the best performance with the lowest error (mean absolute error, root mean squared eror, and mean absolute percent error) and highest coefficient of determination (R2), Kling-Gupta efficiency, and index of agreement between the experimental and predicted shear capacities. Moreover, the accuracy of the proposed xgBoost model was compared with that of the available code and guideline equations and resulted in a superior prediction capability.