Predicting of load capacity of concrete columns confined with FRP bars and subjected to axial compression at different eccentricity levels

In this study, two new models were developed to predict the peak axial capacity of reinforced concrete (RC) compressive members having fiber-reinforced polymer (FRP) bars at different eccentricity levels (e/h = 0 and e/h ranges from 0.08 to 1) using two distinct methods: the general regression method and the eXtreme Gradient Boosting (XGBoost) algorithm. These models were developed based on a wide range dataset comprising tests data of 308 FRP-reinforced concrete samples compiled from the existing literature. Besides, the efficiency and accuracy of the proposed models were assessed using five statistical indicators namely, coefficient of determination (R2), root mean square error (RMSE), mean absolute error (MAE), average absolute error (AAE), standard deviation (SD), and were equated with design codes and previously proposed formulas in the literature. The findings demonstrate that the suggested estimation models were suitable for capturing the axial capacity of FRP-RC compressive members. Particularly, the XGBoost model exhibited outstanding performance with a high R2 value of 0.98 and minimal RMSE, MAE, AAE and SD values of 259.05 kN, 144.36 kN, 0.11, and 0.14 respectively, indicating excellent efficiency and accuracy compared to both the empirical model proposed and other existing models. This outcome highlights the ability of machine learning models to estimate the axial capacity of FRP-RC compressive members. Consequently, the XGBoost model offers a viable alternative method to empirical models for design applications.