Explainable Geospatial Artificial Intelligence for Malaria Risk Forecasting: A Multi-Indicator Framework.

The datasets cover Uganda (2010–2020) and Zimbabwe (2015–2022) and were derived from multiple satellite and epidemiological data sources including the Malaria Atlas Project (MAP), CHIRPS, GIMMS-3G+ NDVI, ERA5 reanalysis, and NASADEM elevation data. The following files are included:

spatio_temporal_malaria_indices.csv — Annual spatiotemporal malaria epidemiological indices for Uganda, including parasite rate, incidence rate, mortality rate, and insecticide-treated net access and usage at 5km spatial resolution. These are the raw input indicators used to derive the composite Malaria Risk Index (MRI) via Principal Component Analysis (PCA).

malaria_risk_index.csv — Annual composite MRI values for Uganda (2010–2020) derived from PCA applied to the five epidemiological indicators. Each row represents a unique location (latitude, longitude) and year combination.

malaria_risk_index_monthly.csv — Monthly downscaled MRI values for Uganda (2010–2020) estimated using a state-space dynamic latent variable model driven by monthly NDVI, precipitation, and temperature covariates.

ecological_predictors_with_mri.csv — Merged dataset combining monthly MRI values with satellite-derived environmental covariates (NDVI, precipitation, temperature, elevation, and engineered seasonality features) for Uganda. This is the primary training dataset for the XGBoost and LSTM models.

ZWE_combined.csv — Merged dataset of monthly environmental covariates and MRI values for Zimbabwe (2015–2022), used for cross-country model evaluation and fine-tuning.

ZWE_malaria_risk_index_monthly.csv — Monthly downscaled MRI values for Zimbabwe (2015–2022) generated using the same state-space downscaling methodology applied to Uganda.

ZWE_malaria_risk_index_yearly.csv — Annual composite MRI values for Zimbabwe (2015–2022) derived from PCA applied to the five epidemiological indicators at 5km spatial resolution.