Abstract: Objective To develop risk prediction models for postpartum stress urinary incontinence(PPSUI)using five machine learning algorithms based on the latest PPSUI definition, identify the optimal model, and evaluate its clinical applicability. Methods This study adopted a prospective design and included 1,208 postpartum women. Data were collected from questionnaires and electronic medical records, and the dataset was randomly divided into a training set and a test set in an 8∶2 ratio. Feature selection was performed using univariate analysis and the random forest algorithm. Five PPSUI prediction models were developed based on Logistic regression, decision tree, random forest, support vector machine, and extreme gradient boosting(XGBoost), with hyperparameter optimization conducted through grid search. Model training and validation were performed using the bootstrap method and ten-fold cross-validation to enhance stability and generalizability. The models were comprehensively evaluated in terms of classification performance, clinical applicability, and predictive reliability to identify the optimal prediction model. Results XGBoost was identified as the optimal prediction model, achieving the highest AUC in the test set(AUC=0.993, 95%CI: 0.985-0.998, P<0.01). Decision curve analysis showed that XGBoost provided the highest net benefit within the intermediate threshold range, while the calibration curve was closest to the ideal state, indicating superior predictive reliability and significant clinical applicability. Conclusion XGBoost is suitable for early screening and risk assessment of high-risk PPSUI populations, providing a scientific basis for precision medicine and postpartum health management.

Key words: Stress urinary incontinence, Postpartum, Risk prediction, Machine learning, Predictive model