Machine-learning algorithm to predict home delivery after antenatal care visit among reproductive age women in East Africa.

Background: Maternal and child health remains a global public health issue, particularly in low- and middle-income countries where maternal and child mortality are extremely high. The World Health Organization estimates that close to 287,000 women die annually due to pregnancy and childbirth complications, and the majority of these deaths occur where skilled birth attendants are not readily available. Reducing the prevalence of home delivery is a key strategy for lowering the maternal mortality rate. Although several studies have explored home delivery and antenatal care (ANC) utilization independently, limited evidence exists on predicting home delivery after ANC visits using machine-learning approaches in East Africa. Methods: This study utilized a community-based, cross-sectional design with data from the most recent Demographic and Health Surveys conducted between 2011 and 2021 in 12 countries in East Africa countries. A total weighted sample of 44,123 women was analyzed using Python version 3.11. Nine supervised machine-learning algorithms were applied, following Yufeng Guo's steps for supervised learning. The random forest (RF) model, selected as the best-performing algorithm, was used to predict home delivery after ANC visits. A SHapley Additive exPlanations analysis was conducted to identify key predictors influencing home delivery decisions. Results: Home delivery after ANC visits was most prevalent in Malawi (17.88%), Uganda (15.38%), and Kenya (11.3%), and was low in Comoros (2.38%). Living in rural areas and late ANC initiation (second trimester) increased the likelihood of home delivery. In contrast, factors such as higher household income, husband's level of primary and secondary education, contraceptive use, shorter birth intervals, absence of distance-related barriers to healthcare, and attending more than four ANC visits were associated with a lower likelihood of home delivery. Conclusion: The study demonstrates that home delivery after ANC visits was high. The RF machine-learning algorithm effectively predicts home delivery. To reduce home deliveries, efforts should improve early ANC initiation, enhance healthcare quality, and expand facility-based services. Policymakers should prioritize increasing health facility accessibility, promoting media-based health education, and addressing financial barriers for women with low incomes. Strengthening these areas is crucial for improving maternal and neonatal health outcomes in East Africa.