Machine learning approaches to predict maternal anemia and Identifying associated factors among pregnant women in Sub-Saharan Africa

Anemia remains a major global public health problem, affecting an estimated 1.93 billion people worldwide. Pregnant women in sub-Saharan Africa (SSA) are disproportionately affected, placing them at greater risk of adverse maternal and neonatal outcomes. This study used machine learning to predict anemia and identify key predictors among pregnant women in SSA. We analyzed the most recent Demographic and Health Survey (DHS) data from 24 SSA countries, including a weighted sample of 14,569 pregnant women. Data were prepared in SPSS version 27 and analyzed in Python version 3.12. An Extreme Gradient Boosting (XGBoost) classifier was developed to predict anemia, while Shapley Additive Explanations (SHAP) identified influential predictors and improved model interpretability. Model performance was assessed using accuracy, recall, F1 score, and area under the receiver operating characteristic curve (AUC). The XGBoost model achieved 87% accuracy, 85% recall, a 73% F1 score, and an AUC of 95%, indicating excellent predictive performance. SHAP analysis identified unimproved water sources, lack of antenatal care visits, absence of mobile phone ownership, older maternal age, financial barriers to healthcare, cigarette smoking, khat chewing, limited media exposure, urban residence, home delivery, and use of unimproved cooking fuel as the most influential predictors of anemia. These findings demonstrate that machine learning can accurately identify pregnant women at high risk of anemia using routinely collected sociodemographic, behavioral, and maternal health data. Integrating ML-based prediction models into antenatal care and digital maternal health platforms could support timely risk identification, personalized interventions, and targeted strategies to reduce maternal anemia across SSA.