Prediction and Interpretability Analysis of RC Column Yield Displacement Based on Bayesian Optimization Ensemble Learning

Aiming at the problems of high cost and large error of traditional RC column yield displacement calculation methods, an ensemble learning algorithm is adopted to predict the yield displacement of RC columns. Firstly, 160 sets of RC columns yield displacement proposed pseudo-static testing data are collected, the data are Z-score standardized and Spearman correlation analysis is applied to analyze the features affecting the influence of yield displacement (including, \(\eta \) , \(\lambda \) , H, and b, etc.). Then establish RF (Random Forest), AdaBoost (Adaptive Boosting), GBRT (Gradient Boosting Regression Tree), XGBoost (Extreme Gradient Boosting), and CatBoost five yielding displacement prediction models and Bayesian optimize of the hyperparameters of the models. Finally, the evaluation metrics of each prediction model are compared and analyzed with the theoretical formulation of yield displacement, and the ensemble learning prediction models of yield displacement are explained using Shapley additive explanations (SHAP). The comparative results are \({R}^{2}\) , RMSE, and MAE of the best ensemble learning prediction model are 0.98, 2.05, and 1.51 respectively, which are better than the theoretical formulae, indicating that the ensemble learning prediction model has a better fitting ability with higher accuracy. Explainable results are the larger SHAP values for column length, shear span ratio, longitudinal reinforcement yield strength, hoop reinforcement yield strength and axial compression ratio, indicating that these features have a greater influence on the ensemble learning prediction model for yield displacement. The proposed model provides a practical tool for seismic design and assessment of RC columns.