A data-driven prediction method for shear bearing capacity of flexural-shear type RC column based on PCC-XGB-SHAP

The calculation of shear bearing capacity of RC columns is crucial for seismic design and safety assessment. However, due to the complex shear mechanisms of RC columns under seismic loading and the multitude of influencing factors, there is a lack of theoretically rigorous and comprehensive shear bearing capacity models internationally and domestically. This paper aims to establish a unified, data-driven prediction model for the shear bearing capacity of flexure-shear RC columns and achieve rapid predictions. Based on an exhaustive search strategy, the study identifies key feature parameters affecting flexure-shear RC columns. Pearson correlation analysis is used to examine the correlation between target shear bearing capacity and each influencing parameter. An XGB algorithm is employed to construct a data-driven prediction model for shear bearing capacity. The model undergoes quantitative evaluation through 5-fold cross-validation and is compared with DT, SVM and RF models. Results indicate that the XGB model has the highest accuracy and robust generalization ability, making it the most suitable ML method for shear capacity prediction. Based on the prediction data of the optimized model, the SHAP method is applied to address the model interpretability issue in the shear bearing capacity prediction of flexure-shear RC columns. Furthermore, by comparing the results of existing empirical calculation formulas and the ML model, the established model is validated, demonstrating superior predictive performance. This study provides valuable references for the design and safety assessment of RC columns.