Personalized machine learning model for predicting successful vaginal delivery during the active phase of labor.

This study asks whether a personalized, explainable machine learning model can predict successful vaginal delivery in real time during the active phase of labor using both static maternal characteristics and dynamic intrapartum data. Using 70,000 active-phase labor records from a tertiary hospital (2010-2021), the authors found that a Histogram-Based Gradient Boosting model performed best (ROC AUC 0.87, PR AUC 0.52), with fetal descent rate, cervical opening rate, maternal age, and clinical fetal weight estimate contributing most to predictions, as revealed by SHAP analysis.