onsdev2001/sanadtn

AI-powered preventive health system for Tunisia — chronic disease risk detection using ML ensemble (RF + XGBoost), SHAP explainability, and INS Tunisia data integration. # Sanad AI — سند > **National Preventive Health System for Tunisia** — An intelligent platform for early detection of chronic diseases (diabetes, cardiovascular disease), adapted to the Tunisian context by integrating socio-environmental INS data for each governorate (wilaya). --- ## 🎯 Features - **ML Model** — Random Forest + XGBoost ensemble (81.5% recall on positive class) - **Tunisia Layer** — 12 real INS indicators across all 24 wilayas - **SHAP Explainability** — top 3 determining risk factors per prediction - **Ministry Dashboard** — interactive choropleth risk map of Tunisia - **React Interface** — citizen and nurse modes - **JWT Authentication** — 3 roles (nurse, minister, admin) - **PostgreSQL** — persistent storage of predictions --- ## Screenshots ### Ministry Dashboard — National KPIs ### Interactive Risk Map by Wilaya ### Patient Risk Simulator ### SHAP Explainability — Top Risk Factors ### Secure JWT Login ### React Interface — Citizen Mode ### React Interface — Nurse Mode --- ## Architecture ``` sanad/ ├── api/ → FastAPI backend (/predict, /heatmap, /login) ├── dashboard/ → Streamlit Ministry dashboard ├── data/geo/ → Tunisia INS data (ins_tunisia.py) ├── db/ → PostgreSQL schema and functions ├── frontend/ → React interface (citizen + nurse) ├── model/ → Trained models (.pkl) ├── notebooks/ → Jupyter notebooks (Sprint 1, 2, 3) └── docs/ → Project report (PDF) +presentation ``` --- ## How It Works ``` ┌──────────────┐ ┌─────────────────┐ ┌──────────────────┐ │ Patient │────▶│ Feature Vector │────▶│ RF + XGBoost │ │ Input │ │ + Tunisia Layer │ │ Ensemble │ └──────────────┘ │ + INS Data │ └────────┬─────────┘ └─────────────────┘ │ ▼ ┌────────────────┐ ┌──────────────┐ │ SHAP │◀───│ Risk Score │ │ Explanation │ │ + Level │ └────────────────┘ └──────────────┘ ``` The mod …