aeroprice

A fullstack flight fare prediction app. Enter your airline, route, date, and departure time — the ML model spits out an estimated ticket price in seconds.

Live demo: https://aeroprice-v2.vercel.app/

What it does

• Takes flight details as input — airline, source, destination, number of stops, flight duration, journey date, and departure time
• Derives additional features from that input: day of week, time-of-day category (morning/afternoon/evening/night), arrival hour/minute
• Feeds a 16-feature vector into a trained scikit-learn model and returns a predicted fare
• Frontend has a dark space-themed UI with animated starfield, aurora blobs, and a flying plane animation that triggers on prediction

---

Tech stack

Frontend — React 19, Create React App, deployed on Vercel
Backend — Python, FastAPI, Uvicorn
ML — scikit-learn model trained on Indian domestic flight data, serialised as a .pkl file
Data processing — NumPy, Pandas

---

Project structure

aeroprice-fullstack/
├── backend/
│   ├── main.py                   # FastAPI app + /predict endpoint
│   ├── model.py                  # Loads the pickled model
│   ├── flight_price_model.pkl    # Trained ML model
│   └── requirements.txt
└── frontend/
    ├── public/
    └── src/
        ├── App.js                # Full UI — form, animations, prediction result
        ├── App.css
        └── index.js

---

How the prediction works

The model takes 16 numerical features built from the user's input:

Index	Feature
0	Airline (label encoded)
1	Destination (label encoded)
2	Number of stops
3	Journey day
4	Journey month
5	Departure hour
6	Departure minute
7	Arrival hour (computed from duration)
8	Arrival minute
9	Duration hours
10	Duration minutes
11	Day of week
12	Time category (0=morning, 1=afternoon, 2=evening, 3=night)
13–15	Source one-hot encoded (Bangalore, Kolkata, Delhi, Chennai, Mumbai)

Supported airlines: IndiGo, Air India, Jet Airways, SpiceJet, Vistara
Supported sources: Bangalore, Kolkata, Delhi, Chennai, Mumbai
Supported destinations: Cochin, Delhi, Hyderabad, Kolkata

---

Running locally

Backend

Requires Python 3.8+.

cd backend
pip install -r requirements.txt
uvicorn main:app --reload

The API will be running at http://localhost:8000.

Test it:

curl -X POST http://localhost:8000/predict \
  -H "Content-Type: application/json" \
  -d '{
    "airline": "IndiGo",
    "source": "Banglore",
    "destination": "Delhi",
    "stops": 0,
    "duration": 150,
    "journey_date": "2026-06-15",
    "departure_time": "06:30"
  }'

Frontend

Requires Node.js 16+.

cd frontend
npm install
npm start

The app will be at http://localhost:3000. By default it points to http://localhost:8000 for the backend — update the API URL in App.js if you're deploying the backend elsewhere.

---

API

POST /predict

Request body:

{
  "airline": "IndiGo",
  "source": "Banglore",
  "destination": "Delhi",
  "stops": 1,
  "duration": 180,
  "journey_date": "2026-06-15",
  "departure_time": "08:45"
}

Response:

{
  "predicted_price": 4823.57
}

Validation rules:

• duration must be between 60 and 1000 minutes
• stops must be 0 or more
• journey_date and departure_time are required

GET /

Health check. Returns { "message": "Flight Price API Running" }.

---

Deploying

Backend — works on any platform that supports Python. For Render:

1. Set build command: pip install -r requirements.txt
2. Set start command: uvicorn main:app --host 0.0.0.0 --port $PORT

Frontend — already deployed on Vercel. For a fresh deploy, push to GitHub and import the repo on Vercel. Set REACT_APP_API_URL as an environment variable pointing to your deployed backend URL, then update App.js to use process.env.REACT_APP_API_URL.

---

Notes

• The .pkl model file is ~46MB — GitHub has a 100MB file size limit so you're fine, but if the file ever grows past that you'll need Git LFS.
• The model was trained on historical Indian domestic flight data. Predictions outside the trained routes and airlines will fall back to default encoding (index 0) and may be less accurate.
• duration input is in total minutes — the backend splits it into hours and minutes before passing to the model.