Machine learning approaches to predict health facility delivery among reproductive-age women in East Africa

Delivery at health facilities offer access to emergency obstetric care and guarantee handling of problems like eclampsia, postpartum hemorrhage, and obstructed labor. Using machine learning to predict delivery at a health facility supports automated decision-making, enhances classification accuracy, and scales well when dealing with large and complex datasets. The aim of this research was to predict health facility delivery among women of reproductive age in East Africa by applying a machine-learning approach. A cross-sectional study design was used, drawing on Demographic and Health Survey (DHS) datasets from 11 East African countries collected between 2015 and 2024. Data preprocessing involved cleaning, transformation, and integration, along with one-hot encoding, 80/20% data splitting, and 10-fold stratified cross-validation. Four machine-learning algorithms were assessed, such as random forest, decision tree, k-nearest neighbors, and Naïve Bayes. The random forest classifier achieved the highest performance, yielding an area under the curve (AUC) of 84.8% and an accuracy of 82.9%. Shapley Additive exPlanation (SHAP)-based feature importance analysis identified wealth index, women’s and husbands’ educational status, number of children, place of residence, distance to health facility, and frequency of Antenatal Care (ANC) visits as the strongest predictors of health facility delivery. These results offer actionable insights for planners, Non-Governmental Organizations (NGOs), and policymakers to design targeted and culturally appropriate interventions that improve equitable access to maternal health services. Future studies should test advanced models such as ensemble and deep learning, integrate geospatial and health system data, and develop secure real-time predictive tools within national health platforms.