Abstract: To obtain accurate prediction model for effective stiffness of RC rectangular hollow piers, a dataset containing 131 samples with bending failure as the main failure mode was established. The applicability of existing effective stiffness models to rectangular hollow piers was analyzed. Six machine learning models for effective stiffness prediction, including support vector regression (SVR), random forest regression (RFR), extreme random tree regression (ERTR), gradient boosting tree regression (GBTR), extreme gradient boosting tree regression (XGBR), and voting ensemble regression (VOTR), were constructed with 13 features as input parameters. The prediction performance of existing models and machine learning models was evaluated by using 131 samples. And the SHAP method was used to explain the XGBR model. The research showed that, except for Wang's and Wei's models, existing formulas significantly overestimated the effective stiffness of rectangular hollow piers on mean value of statistical meaning, and the coefficients of variation of each model were relatively large. Compared with existing models, machine learning algorithms had great advantages, and even the SVR model with the lowest prediction performance had higher prediction accuracy than all existing models. Among single-ensemble learning models, the XGBR model had the best prediction performance. The proposed double-ensemble VOTR model further improved the prediction performance and had the highest prediction accuracy. The root-mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and coefficient of determination (R2) of VOTR model on the complete data set were 1.3%, 0.7%, 3.4%, and 0.990, respectively; and the average and coefficient of variation of the ratio of predicted value to test value were 1.01 and 0.07, respectively. The VOTR model obtained the most stable, safe, and accurate prediction results, which was far superior to existing models. The SHAP method can explain the model prediction results from both the global and individual levels. For XGBR model, the top five features in importance order were shear span ratio L/H, longitudinal reinforcement ratio ρl, axial compression ratio n, longitudinal reinforcement yield strength fyl, and material geometric parameter m (i.e. fyldb /Lfc ), which were helpful for improving the physical analysis model of effective stiffness of RC rectangular hollow piers.