akbempah1/malaria-prediction-africa

Predicting malaria incidence across Sub-Saharan Africa using climate and demographic variables — regression modeling with WHO and World Bank data. # Malaria Prediction in Sub-Saharan Africa **Predicting malaria incidence across 44 African countries using GDP, rural population, and health expenditure — Random Forest regression achieving R²=0.789 with WHO and World Bank live data.** > Author: Afriyie Karikari Bempah, PharmD | LinkedIn | GitHub --- ## Overview This project builds a regression model to predict malaria incidence across Sub-Saharan Africa using socioeconomic and health system variables. Data is pulled live from two public APIs — WHO GHO and World Bank — and merged into a single panel dataset covering 44 countries from 2000 to 2023. --- ## Key Findings | Finding | Implication | |---|---| | **Africa malaria incidence fell 40% since 2000** | Bed nets, artemisinin treatment, and donor funding are measurably working | | **Random Forest R²=0.789 vs Linear R²=0.474** | Non-linear socioeconomic interactions drive malaria burden | | **GDP per capita is the strongest predictor** | Poverty is the primary structural driver of malaria risk | | **Niger underpredicted** | Extreme seasonality not captured by annual socioeconomic features | | **Rwanda and Ghana outperform predictions** | Community health programs deliver outcomes beyond GDP expectations | --- ## Model Performance | Model | R² | RMSE | |---|---|---| | Linear Regression | 0.474 | 1.313 | | Ridge Regression | 0.474 | 1.313 | | **Random Forest (Tuned)** | **0.785** | **0.840** | --- ## Technical Approach - **Multi-API data pipeline** — WHO GHO + World Bank merged on country code and year - **Log transformation** — applied to malaria incidence and GDP for linearity - **Three model comparison** — Linear, Ridge, Random Forest - **GridSearchCV tuning** — 12 parameter combinations, 5-fold CV - **Country-level 2023 predictions** — actual vs predicted for all 44 countries --- ## Skills Demonstrated - Multi-source API integration and data merging - Log transformation for skewed regression targets - Regression modeling (Linear, Ridge, Random F …