Isaackjoshua/Afya_Predict

AFYA-PREDICT (Afya = "Health" in Swahili), a modular, extensible AI disease-outbreak prediction platform designed for Tanzania and scalable across Africa. # AFYA-PREDICT **AI-Powered Fusion Yield Analysis for Predictive Disease Intelligence in East Africa & Beyond** *Afya* is Swahili for "health". AFYA-PREDICT is a free, open-source, explainable outbreak-**forecasting** platform. It fuses satellite climate, mobility, population, air-quality, WASH and routine health-surveillance streams onto a common `district × epidemiological week` grid and predicts **where** and **when** disease risk will rise — typically 4–12 weeks before cases reach facility reports. Built for Tanzania first, designed to scale across Africa. MIT licensed. --- ## Why another system? | Existing gap | AFYA-PREDICT | |---|---| | Reactive surveillance (DHIS2/eIDSR, HealthMap, ProMED) | Pre-outbreak forecasting from upstream climate and mobility drivers | | Single-source fragility (Google Flu Trends) | ≥3 fused sources per disease; one feed failing degrades quality, not availability | | Never retrained (GFT 2008–2015) | Page-Hinkley + ADWIN drift detection with automatic, holdout-gated refits | | Black box (BlueDot) | SHAP attribution and a plain-language explanation on **every** prediction | | Siloed human/animal/environment data | One Health schema feeding a single engine | | Costly and proprietary | Free data sources, commodity hardware, no vendor lock-in | | No spatial spread | Mobility-weighted diffusion naming the source districts, gravity-model fallback | | Single disease | Plugin architecture — a new disease is a YAML file and a short class | | Alerts without actions | Every alert carries costed, timeboxed, owned recommendations | | Rarely externally validated | Walk-forward CV; three naive baselines a model must beat to deploy | | Digital-divide bias | Primary signals are satellite + DHIS2, never search or social | | No feedback loop | Interventions logged, impact triangulated, contaminated weeks down-weighted | --- ## Quick start ```bash git clone github.com cd Afya_Predict cp .env.example .env …