Abstract: Numerous factors influence the punching shear failure of RC slab-column connections under the combined action of bending moment and shear force. Traditional load-bearing capacity models based on empirical or semi-empirical formulas still face the challenge of insufficient accuracy. This paper proposed an interpretable machine learning model for predicting the load-bearing capacity of slab-column connections with shear reinforcement. The model leverages the XGBoost algorithm to learn the implicit mapping relationship between high-dimensional parameters, such as geometric and material properties and the punching shear capacity. The SHAP interpretation method is used to analyze the sensitivity of parameters. A punching shear database of RC slab-column connections with shear reinforcement, consisting of 235 specimens, was constructed. The proposed method was compared with other machine learning models and code-based formulas to assess its effectiveness. Results show that the proposed model outperforms other models and code-based formulas in terms of prediction accuracy and stability, with mean, standard deviation and coefficient of variation of the prediction results being 1.00, 0.07 and 7.2%, respectively. Using the SHAP method, the influence of each parameter on the punching shear capacity of slab-column connections was quantified. It is found that the geometric properties of the components dominate the punching shear capacity, with the effective depth d being the most critical variable, followed by the shear reinforcement area within a range of 1.5 d from the column face, which accounts for about 35% of the influence of d, while the influence of material properties is relatively minor.