Magguire/Kenya-Malaria-Risk-Mapping

# Kenya Malaria Hotspot-Shift Mapping Identify how malaria risk hotspots in Kenya have shifted over time, at county resolution, using a pooled Random Forest model and SHAP interpretation. ## Overview The ideal version of this analysis would compare Kenya's 2015 and 2020 Malaria Indicator Survey (KMIS) rounds directly; two matched, structured national surveys. That data is gated behind DHS Program registration (free, but approval takes 1–3 business days and can't be automated in Colab). **This notebook uses a proxy instead**: the openly-downloadable Malaria Atlas Project (MAP) parasite-rate survey database, split into two multi-year windows around Kenya's **2013 devolution of health services** a real, policy-relevant cut point, since that's when malaria control planning shifted to being organized around counties. This is a weaker design than the real KMIS comparison, for one specific reason: MAP's survey points are opportunistic, different research studies surveyed different locations in different years, so an apparent change between the two windows partly reflects *which places happened to get surveyed when*, not only true epidemiological change. See Limitations below. ## Data Sources | Source | Used for | Access | |---|---|---| | Malaria Atlas Project (MAP) `malariaAtlas` R package | PfPR survey points, Kenya county boundaries | Open, no registration | | Google Earth Engine CHIRPS, MODIS, WorldPop | Rainfall, NDVI, land surface temperature, population density covariates | Free, requires registration | | DHS Program KMIS 2015 / 2020 | *Not used here* the ideal ground truth this notebook is a proxy for | Gated, requires registration | ## Methodology 1. **Setup** - install Python packages (`geopandas`, `shap`, `scikit-learn`, `mapclassify`, `contextily`, `xgboost`) 2. **Install R** and the `malariaAtlas` package inside the Colab session 3. **Pull MAP data** for Kenya - PfPR survey points and county (admin1) boundaries 4. **Validate the survey-year distrib …