Abstract:
Introduction: The World Health Organization (WHO) defines stillbirth as a baby weighing at least
1000 g at birth or showing no signs of life at or after 28 weeks of pregnancy. According to literatures
traditional approaches are unable to handle the amount, complexity, and velocity of healthcare big
data and limited uses of advanced machine learning techniques
Objective: This study aims to predict stillbirth and its determinants among gave birth women in
East Africa, using a machine learning algorithm: Evidence from the recent Demographic Health
Survey, 2016-2022/23
Methods: This study used data from a recent demographic health survey of selected East African
countries. A weighted sample of 478,548 women of reproductive age was included in the study.
The data were cleaned and weighed using STATA version 17. A Supervised machine learning
algorithms Were utilized using python software Additionally, to forecast and determine the most
significant predictors of stillbirth, eight supervised machine learning algorithms were used.
Results: The study identified a random forest algorism was the outstanding to predict still birth,
with accuracy 90%, AUC97.4%, recall 92%, precision 88%.
According to a random forest model-based SHAP study, the most significant predictors of stillbirth
in East Africa were no ANC visits, aged respondents, poor wealth, single marital status, respondents
with no formal education, alcohol consumption, and respondents with no work.
Conclusions: Overall, the Random Forest classifier has a higher chance of detecting stillbirth and
determining its causes. The study recommends increasing and giving prior attention to maternal
and child health through health promotion, declaring free home delivery, and improving access and
coverage of maternal health services at healthcare facilities.
