SmartQueue — IEEE Final Project (v3)

Architecture

smartqueue-final/
├── ml-service/     ← Python Flask + GradientBoosting ML (port 8000)
├── backend/        ← Node.js + Express + MongoDB (port 5000)
└── frontend/       ← React.js (port 3000)

---

Step 1 — Start ML Service (Python)

cd ml-service
pip install flask scikit-learn numpy pandas joblib
python app.py

Model auto-loads from model.pkl. Opens at http://localhost:8000

• /predict-slots → ML wait time for all 6 time slots
• /predict → Single slot prediction
• /health → Model accuracy stats (R²=0.9985, MAE=2.32 min)
• /model-report → Full training report

---

Step 2 — Start Backend (Node.js)

cd backend
npm install
npm run dev

Opens at http://localhost:5000. Requires MongoDB running locally.

To make admin account: After registering, open MongoDB Compass:

db.users.updateOne({ email: "admin@gmail.com" }, { $set: { role: "admin" } })

---

Step 3 — Start Frontend (React)

cd frontend
npm install
npm start

Opens at http://localhost:3000

---

AI/ML Features (IEEE Paper)

Feature	Implementation	File
Wait time prediction	GradientBoostingRegressor (n=200, R²=0.9985)	ml-service/train.py
Peak hour heatmap	30-day booking aggregation → 5×6 matrix	backend/utils/helpers.js
AI visit suggestions	ML ranks all 6 slots by predicted wait	ml-service/app.py
Tribonacci priority	Dynamic priority score, grows with wait time	backend/utils/helpers.js
QR token	Unique tokenId per booking, scannable	backend/routes/bookings.js
QR scanner	html5-qrcode camera scanner	frontend/src/components/QRScanner.js
Nearby alternatives	MongoDB query same-category lower queue	backend/routes/businesses.js

ML Model Details

• Algorithm: GradientBoostingRegressor (scikit-learn)
• Training samples: 8000
• Features: slot_hour, day_of_week, queue_length, avg_service_time, is_peak_hour, bookings_30d_slot, elderly_frac
• MAE: 2.32 minutes
• R²: 0.9985
• CV R²: 0.9984