A Full-Stack AI Public Health Decision Support System

📖 Docs · 🚀 Quick Start · 🏗️ Architecture · 🤖 ML Models · 🌐 API Reference

---

🎯 What is FAST?

FAST (Forecasting Aggregate-level Self-harm Trends) is a production-grade, full-stack web application that uses machine learning to forecast national self-harm trends from social media mental health signals (Fear, Anger, Sadness).

⚠️ Research Use Only — Predicts aggregate national trends. NOT for individual clinical risk assessment.

✨ Key Highlights

Feature	Description
🤖 AutoML Engine	Trains & compares 5 models: XGBoost, Random Forest, SVR, Bayesian Ridge, ARIMA
🔐 JWT Authentication	Secure login/register with 24-hour access tokens
📊 Interactive Charts	Plotly.js visualizations — forecasts, signals, feature importance
🎯 Policy Simulator	What-If analysis by adjusting Fear/Anger/Sadness signals
🔍 Explainable AI	Feature importance to understand prediction drivers
📄 Report Generator	Export full markdown analysis reports
🌐 REST API	Full FastAPI backend with Swagger UI at /docs
💾 Persistent Storage	SQLite database for users, datasets, and models

---

🛠️ Tech Stack

Layer	Technology
Frontend	React 18, Vite, React Router v6, Plotly.js, Axios
Backend	FastAPI, Uvicorn, SQLAlchemy, Pydantic v2
Authentication	JWT (python-jose), bcrypt (passlib)
Machine Learning	Scikit-learn, XGBoost, Statsmodels, SHAP
Data	Pandas, NumPy, GeoPandas
Database	SQLite (SQLAlchemy ORM)
Deployment	Render.com, Railway (config included)

---

🚀 Quick Start

Prerequisites

• Python 3.11+
• Node.js 18+

1. Clone

git clone https://github.com/6673csm/FAST_Dashboard.git
cd FAST_Dashboard

2. Start the Backend (FastAPI)

cd backend
pip install -r requirements.txt
python -m uvicorn main:app --reload --port 8000

• API running at: http://localhost:8000
• Swagger docs at: http://localhost:8000/docs

3. Start the Frontend (React)

cd frontend
npm install
npm run dev

• App running at: http://localhost:5173

4. (Optional) Run Original Streamlit App

pip install -r requirements.txt
streamlit run app.py

• Streamlit app at: http://localhost:8501

---

🏗️ Architecture

FAST_Dashboard/
│
├── backend/                    ← FastAPI REST API
│   ├── main.py                 ← App entry, CORS, routers
│   ├── database.py             ← SQLite + SQLAlchemy ORM
│   ├── auth.py                 ← JWT authentication
│   ├── schemas.py              ← Pydantic request/response models
│   └── routers/
│       ├── auth.py             ← POST /api/auth/register, /login
│       ├── data.py             ← POST /api/data/upload, GET /list
│       ├── models.py           ← POST /api/models/train
│       ├── forecast.py         ← POST /api/forecast/run
│       ├── simulator.py        ← POST /api/simulator/run
│       └── reports.py          ← GET /api/reports/generate/{id}
│
├── frontend/                   ← React + Vite SPA
│   └── src/
│       ├── pages/              ← 10 page components
│       ├── components/         ← Sidebar, MetricCard, Spinner
│       ├── context/            ← AuthContext (JWT state)
│       └── api/client.js       ← Axios with JWT interceptor
│
├── modules/                    ← Original Streamlit ML modules
├── pages/                      ← Original Streamlit pages
├── app.py                      ← Original Streamlit entry
├── render.yaml                 ← Render.com deployment config
├── Procfile                    ← Railway/Heroku deployment
└── .gitignore

---

🌐 API Reference

Method	Endpoint	Auth	Description
POST	/api/auth/register	❌	Create new user account
POST	/api/auth/login	❌	Get JWT access token
GET	/api/auth/me	✅	Get current user
POST	/api/data/upload	✅	Upload & auto-clean CSV
GET	/api/data/list	✅	List user datasets
GET	/api/data/{id}/summary	✅	Dataset stats + preview
POST	/api/models/train	✅	Train 5 AutoML models
GET	/api/models/list/{id}	✅	Model leaderboard
GET	/api/models/feature-importance/...	✅	Feature importance
POST	/api/forecast/run	✅	Generate future forecast
POST	/api/simulator/run	✅	What-If policy scenario
GET	/api/reports/generate/{id}	✅	Export analysis report

📖 Full interactive docs at http://localhost:8000/docs (Swagger UI)

---

🤖 ML Models

Model	Type	Best For
XGBoost	Gradient Boosting	High accuracy, handles non-linearity
Random Forest	Ensemble	Robust, handles overfitting
SVR	Support Vector	Small datasets, kernel tricks
Bayesian Ridge	Linear	Uncertainty quantification
ARIMA	Time Series	Pure time-series baseline

Feature Engineering

Automatically creates:

• Lag features: 1, 7, 14, 30-day lags for each signal
• Rolling stats: 7-day and 30-day moving averages & std deviations
• Evaluation metrics: MAE, RMSE, R², MAPE

Sample Results (on sample data)

Model	MAE	RMSE	R²	MAPE
XGBoost	2.14	2.67	0.94	5.23%
Random Forest	2.38	2.91	0.92	6.12%
Bayesian Ridge	2.67	3.15	0.89	7.45%
SVR	2.89	3.40	0.87	8.21%
ARIMA	3.12	3.65	0.84	9.34%

---

🎯 Data Format

date,me-fea,me-ang,me-sad,gh-death,gh-injure
2023-01-01,0.45,0.32,0.56,12.3,145.2
2023-01-02,0.48,0.35,0.58,13.1,148.5

Column	Description	Range
date	Date of observation	YYYY-MM-DD
me-fea	Fear signal (social media)	0.0 – 1.0
me-ang	Anger signal (social media)	0.0 – 1.0
me-sad	Sadness signal (social media)	0.0 – 1.0
gh-death	Self-harm death rate (per 100k)	≥ 0
gh-injure	Self-harm injury rate (per 100k)	≥ 0

---

☁️ Deploy to Render (Free)

1. Push to GitHub (already done!)
2. Go to render.com → New Blueprint
3. Connect your 6673csm/FAST_Dashboard repository
4. Render reads render.yaml → auto-deploys both API and frontend

---

⚠️ Ethical Considerations

✅ Intended Use	❌ Not For
National trend forecasting	Individual risk assessment
Public health policy planning	Clinical diagnosis
Intervention impact testing	Targeting individuals
Academic research	Medical advice

---

📝 License

MIT License — Free for educational and research use.

---

👨‍💻 Author

Sujay · B.Tech CSE 2026 · India

---

Built with passion for Public Health AI 🧠