Machine LearningEDHSSHAPSpatial Analysis
Institutional Delivery Prediction (EDHS 2024-25)
ML pipeline predicting skilled birth attendance using XGBoost vs. logistic regression, with SHAP explainability and spatial residual mapping.
Built an end-to-end machine learning pipeline on the EDHS 2024-25 Births Recode dataset to predict institutional delivery. Compared XGBoost against logistic regression, reaching a test AUC of 0.88. SHAP analysis identified ANC visits, urban residence, wealth, and travel time as the strongest predictors. A spatial hotspot analysis of out-of-fold residuals highlighted geographic clusters of under- and over-prediction, pointing to where local health system factors matter most.