Conventional urodynamic testing cannot meet the needs of continuous monitoring and real-time pressure warning in smart urinary catheters. A multidimensional and explainable method is proposed to predict end-of-storage bladder pressure. A public urodynamic dataset containing 506 patient records was used. Eight input variables were constructed from pressure, urodynamic function, and demographic characteristics. Random forest, support vector regression, LightGBM, and XGBoost models were trained and compared. Shapley additive explanations (SHAP) analysis and ablation experiments were performed to evaluate feature contributions. Model size was also considered in the performance–complexity analysis. The results show that XGBoost achieves the best overall predictive performance. Pressure fluctuations and pressure levels during bladder filling are identified as the main predictive factors. LightGBM provides a favorable balance between predictive performance and model size. The proposed method provides an algorithmic basis for pressure warning and embedded deployment in smart urinary catheter systems.