nyamilanzi/malawi-maternal-health-clustering

Comparative GMM clustering analysis of maternal health care utilisation patterns from the 2015 and 2024 Malawi DHS # Temporal Patterns of Maternal Health Care Utilisation in Malawi ## A Comparative Unsupervised Learning Analysis of the 2015 and 2024 Demographic and Health Surveys --- ## Overview This repository contains all analysis code, results, and figures for a comparative study of maternal health care utilisation patterns in Malawi, using nationally representative data from the 2015 and 2024 Malawi Demographic and Health Surveys (MDHS). The study applies Gaussian Mixture Model (GMM) clustering to identify distinct utilisation patterns across the antenatal, delivery, and postnatal care continuum, and tracks how these patterns changed over a decade of health system investment. **Key findings:** - A four-cluster solution (2015) and three-cluster solution (2024) best described utilisation heterogeneity - The high-coverage cluster nearly doubled (24.2% → 46.7%), driven by a 31.5 percentage point rise in early ANC initiation - The minimal utilisation cluster (8.8% in 2015) had no identifiable analogue in 2024 - Sociodemographic gradients narrowed across all subgroups, but adolescent women and the poorest quintile retain persistent gaps --- ## Repository Structure ``` . ├── 00_run_all.py # Master script: runs full pipeline in sequence ├── 01_data_harmonization.py # Harmonises 2015 and 2024 MDHS datasets ├── 02_clustering_by_year.py # GMM clustering per survey year (K selection) ├── 03_comparative_analysis.py # Cross-year cluster alignment and temporal comparison ├── 04_subgroup_analysis.py # Cluster distributions by residence, wealth, age, education ├── 05_predictors_membership.py # Multinomial logit + Random Forest predictors ├── 06_visualizations.py # Publication-ready figures (Fig 1–5) ├── build_table1_and_figs.py # Table 1 (demographics) + Figure 2 cluster profiles ├── generate_scatter.py # Figure 6: PCA scatter plot ├── requirements.txt # Python dependencies │ ├── re …