Jeremy-K-coder/diabetes-hypertension-medication-adherence

This project builds and evaluates a classical machine learning pipeline to predict medication non-adherence among diabetic and hypertensive patients in Zimbabwe, using pharmacy refill records and patient-level insurance data from Cimas Medical Aid Society, one of Zimbabwe's largest health insurance providers. ## **PREDICTING MEDICATION NON-ADHERENCE IN DIABETES AND HYPERTENSION PATIENTS USING MACHINE LEARNING** ### Individual Project: Classical Machine Learning for Health Data #### NAME: **`Kirunda Jeremy Menya`** #### REG NO.: **`2025/HD07/25995U`** #### STUDENT NO.: **`2500725995`** *Makerere University | June 2026* *Medication adherence rate breakdown by condition and insurance tier across the study cohort (Zimbabwe, 2022).* --- ## Contents 1. Project Overview 2. Motivation & Significance 3. Research Gap & Originality 4. Clinical Background 5. Dataset 6. Repository Structure 7. Methodology 8. Experimental Design 9. Results 10. Explainability (SHAP) 11. Setup and Usage 12. Ethical Considerations 13. References --- ## Project Overview Non-communicable diseases (NCDs), chiefly Type 2 Diabetes and Hypertension, are the fastest-growing cause of death across Sub-Saharan Africa, now accounting for over 37% of adult mortality in the region. Yet medication adherence rates among NCD patients in Zimbabwe and similar settings remain below 50%, far lower than in high-income countries. When patients do not take their medications consistently, disease progresses silently, complications compound, hospitalisations increase, and healthcare systems that are already under-resourced bear an even greater burden. This project builds and evaluates a classical machine learning pipeline to **predict medication non-adherence** among diabetic and hypertensive patients in Zimbabwe, using pharmacy refill records and patient-level insurance data from **Cimas Medical Aid Society**, one of Zimbabwe's largest health insurance providers. The dataset is real, prospectively collected, and sourced from a published academic study (Kanyongo et al., 2024), making the findings directly relevant to clinical decision-making in a genuine Sub-Saharan African health system context. Beyond standard prediction, this project introduces two original analytical contributions: 1. **Feature group comparison ex …