zda25m006-ship-it/zanzibar-malaria-GNN

# Zanzibar Malaria Importation Prediction with Graph Neural Networks > **Extends:** Muller et al. (2025). *Integrating mobility, travel survey, and malaria case data to understand drivers of malaria importation to Zanzibar, 2022–2023.* Malaria Journal 24:373. doi:10.1186/s12936-025-05605-1 --- ## 📌 Problem Statement Zanzibar has achieved historic reductions in malaria but faces a **"last-mile" elimination challenge**. Its high connectivity with mainland Tanzania — where malaria transmission is significantly higher — causes constant re-introduction of imported cases. **This project predicts** the monthly number and district-level distribution of imported malaria cases in Unguja (Zanzibar's main island) by integrating: 1. **Individual Travel History** — Demographics, travel duration, and destination risk 2. **Environmental Drivers** — Lagged CHIRPS rainfall and NASA POWER temperature 3. **Spatiotemporal Graph Structure** — District adjacency + mobility network + temporal dynamics **Key extension over the paper:** We introduce a **Spatio-Temporal Graph Neural Network (ST-GNN)** pipeline for predictive, leakage-free cross-validated forecasting — something the original paper did not do. --- ## 📂 Datasets | Dataset | Source | Content | |---------|--------|---------| | `ZIM_clinic_data.csv` | ZAMEP (Zanzibar Malaria Elimination Programme) | Anonymized malaria patient records (May 2022 – Dec 2023): travel history, demographics, home district, outcome | | `CHIRPS_rainfall_RAW_ONLY.csv` | CHIRPS | Weekly/Monthly precipitation (mm) for mainland Tanzania regions and Zanzibar. **2–6 week lag built in.** | | `POWER_Regional_Monthly_2022_2023.csv` | NASA POWER | Monthly temperature averages for all districts | > **Malaria importation cases in dataset:** 1,172 of 3,875 total clinic visits (30%) were travel-associated. --- ## 🏆 Key Results > Results below are from **Leave-One-Month-Out Cross-Validation (LOOCV)** on 6 held-out test months (Jul–Dec 2023). Run `pytho …