Machine Learning Mapping of HIV-TB Co-infection across Ghana's 261 Districts

Reproducible pipeline for spatial distribution, determinants, and ML-based risk prediction of HIV-TB co-infection across Ghana's 261 administrative districts (post-2018 Local Governance Act). Combines spatial statistics (Moran's I, LISA, Gi*, GWR), geographically weighted regression, and ensemble machine learning (RandomForest, XGBoost, LightGBM, Stacked). Guan District (Oti Region) added, sharing its parent Krachi East Municipal polygon for spatial-weight purposes (no distinct legacy boundary exists; portfolio-standard structural-gap convention). Canonical statistics (seed=42, 261 districts, recomputed 2026-07-14): global Moran's I=0.472 (HIV-TB co-infection), bivariate Moran's I=0.525 (HIV x TB); 50 LISA High-High clusters, 48 bivariate High-High clusters; GWR R²=0.917; LightGBM 10-fold CV AUC=0.998 (non-spatial, IID folds). Spatial generalisation (leave-one-region-out CV, N=12 regional folds) is materially lower and unstable: LightGBM AUC=0.798 +/- 0.250 -- the large drop and SD reflect genuine spatial autocorrelation in district-level health data, not a bug; the 10-fold number should not be read as the model's out-of-region generalisation performance. If you use this software, please cite it as below.