Abstract: Due to the differences in mechanical properties between steel bars and glass fiber reinforced polymer (GFRP) bars, the axial bearing capacity of GFRP bar-reinforced concrete columns cannot be simply calculated using the methods for reinforced concrete columns. To improve the accuracy of the predictive model for the axial bearing capacity of GFRP bar-reinforced concrete columns, this study used 253 sets of experimental data as the basis for modeling with the extreme gradient boosting (XGBoost) algorithm. Bayesian optimization and Bagging algorithms were employed to optimize the XGBoost algorithm to enhance the model’s predictive accuracy, stability, and training efficiency. The model was evaluated using the coefficient of determination (R²), mean absolute error (MAE), and relative root mean square error (RRSE) and compared with existing predictive models. The study found that Bayesian optimization and Bagging algorithms effectively improved the training efficiency and predictive accuracy of the model. The proposed Bayesian-Bagging-XGBoost model achieved R², MAE, and RRSE values of 0.691 6, 418.162 9, and 0.555 3, respectively, which are significantly better than those of existing predictive models. This model provides a more accurate reference for the engineering application of GFRP bar-reinforced concrete columns.

Key words: Bayesian optimization, XGBoost algorithm, GFRP-reinforced concrete columns, axial bearing capacity, prediction