declerke/Africa-Health-ML

# 🏥 Africa Health Investment Returns **A machine learning pipeline that quantifies the measurable health returns on public investment across 53 African countries — pulling 23 years of World Bank data, training XGBoost models on three health outcomes, and surfacing a Policy Simulator where you adjust Kenya's health-spending sliders and get SHAP-grounded predictions in real time.** | Metric | Value | |---|---| | Countries | 53 African countries | | Time span | 2000 – 2022 (23 years) | | Dataset rows | 1,219 | | Feature variables | 8 | | Target variables | 3 | | Best model R² | 0.945 (Maternal Mortality) | | Pipeline cost | $0 | | App lines of code | 713 (app.py) | | Live dashboard | africa-health-ml.streamlit.… | --- ## 🎯 Project Goal African policymakers and health economists routinely debate whether increasing health spending as a share of GDP actually moves the needle on mortality and life expectancy — or whether the gains are overwhelmed by income levels, sanitation, and education. This project answers that question empirically by training separate XGBoost regressors for Life Expectancy, Under-5 Mortality, and Maternal Mortality, then using SHAP to decompose exactly how much each spending and infrastructure variable contributes to each prediction. The final deliverable is a Policy Simulator: set Kenya's health spending, education spending, water access, and other parameters, and get an immediate, model-backed forecast of expected health outcomes. --- ## 🧬 System Architecture ``` World Bank REST API v2 │ ▼ data_pipeline.py # Fetch → clean → impute → store │ ▼ data/processed/ health.parquet # 1,219 rows, 8 features + 3 targets │ ▼ XGBoost training (in-app) # One model per target variable │ ├── data/models/model_SP_DYN_LE00_IN.json # Life Expectancy ├── data/models/model_SH_DYN_MORT.json # Under-5 Mortality └── data/models/model_SH_STA_MMRT.json # Maternal Mortality │ ▼ SHAP explainer # Per-feature contr …