Accurate prediction of the mode of delivery is critical in maternal care to improve prenatal counseling, optimize clinical decision-making, and reduce maternal and neonatal complications.

This study aims to evaluate and compare the predictive accuracy of AI algorithms in predicting the mode of delivery (vaginal or cesarean) using routinely collected antepartum data from electronic health records (EHRs).

A retrospective dataset of 16,651 pregnancies monitored at St. Mary’s Hospital, London, over a four-year period was utilized. The dataset included 12,639 vaginal deliveries and 4012 unplanned cesarean deliveries, with 92 variables recorded for each patient. Five machine learning algorithms were evaluated: XGBoost, AdaBoost, random forest, decision tree, and multi-layer perceptron (MLP) classifier. A comprehensive feature importance analysis was conducted on the trained models to identify the key predictors influencing the mode of delivery classification.

All five models demonstrated excellent predictive performance, with AdaBoost and XGBoost achieving nearly identical top scores across most metrics: ROC-AUC (90 %), accuracy (89 %), PR-AUC (83 %), and F1 score (88 %) Feature importance analysis highlighted the most predictive factors for mode of delivery. Maternal age demonstrated the highest importance, followed by gravida and maternal height. Additional key contributors included weeks of gestation, - 2-hour plasma glucose level following an oral glucose tolerance test (OGTT), number of previous cesarean sections, and parity.

The findings validate the potential of AI algorithms not only to accurately predict the mode of delivery using antepartum data but also to identify key contributing factors. Integrating such models into clinical decision support systems could enhance prenatal counseling and risk stratification, ultimately contributing to more informed delivery planning and improved maternal and neonatal outcomes.