Reconstructing Inequality in Maternal Health: Imputation-enhanced Machine Learning Models for Global ANC4 Performance

Abstract

The study develops a hybrid composite analysis model that combines imputation and machine learning techniques to predict 527 country-years of national antenatal care coverage with at least four visits (ANC4). The structured missingness of the wealth quintile indicators was addressed through imputations that preserved the distributional characteristics of disadvantaged groups. Five machine-learning algorithms were tested following the imputations. Gradient Boosting achieved the best predictive performance, followed by random forest and XGBoost. Elastic Net, which is more interpretable due to its coefficients, was less predictive but showed significant positive effects on the poorest quartile and rural populations. KNN Regression produced mediocre results and is sensitive to feature scaling. Based on the combined imputation and machine-learning pipeline, it can be concluded that social, economic, and regional disparities exist, with lower-income states and South Asian, Eastern and Southern African regions persistently associated with low ANC4 scores. This research represents a notable innovation, strengthening equity-based maternal health surveillance and providing actionable evidence to advance SDG 3 and SDG 10. It is recommended that global health agencies incorporate various imputation and machine-learning forecasting methods into their routine maternal health monitoring to identify injustices early and allocate resources more effectively.