diaodiallo/ai-predictions-anss: v1.0.0 — AI-Augmented Epidemic Surveillance Pipeline (ANSS - GUINEA)

First public release of the analytical pipeline and DHIS2 native application supporting AI-augmented epidemic surveillance in Guinea. The system predicts epidemic alert threshold exceedances seven days in advance across 19 priority diseases and 38 health districts, by integrating routine DHIS2 early warning surveillance data with NASA POWER climate variables and national census demographics. A LightGBM classifier trained with walk-forward temporal validation achieved an AUC-ROC of 0.902 on the held-out 2024–2025 test set, outperforming the incumbent threshold-based persistence rule by +0.181 AUC. SHAP explanations are computed per prediction and surfaced in the application, so district surveillance officers can inspect and override each alert. Not included Surveillance data are the property of the Agence Nationale de Sécurité Sanitaire (ANSS), Guinea, and are not redistributed here. Per-observation SHAP values are likewise excluded, as they encode district-, disease- and week-level information derived from those data. Both are regenerated by the pipeline from an authorised DHIS2 instance; see docs/REPRODUCE.md. Replication The pipeline is modular and relies exclusively on openly available climate and demographic sources. Deployment in another country requires local retraining and validation, as disease profiles, seasonal transmission dynamics and reporting practices differ substantially across national contexts. Citation Diallo, M. D. (2026). AI-Augmented Epidemic Surveillance: Predicting Epidemic Alert Threshold Exceedances Using DHIS2, Climate and Demographic Data in Guinea. Aix-Marseille Université / ANSS Guinea. doi.org Licence: MIT