Air-India-Flight-Delay-Predictor

Air India Flight Delay Predictor

A production-grade flight delay prediction system for Air India, covering all routes using machine learning.

Features

• Predicts probability of delay >15 minutes
• Predicts expected arrival delay in minutes
• Works for all Air India routes
• Explainable predictions using SHAP
• Python Streamlit frontend

Project Structure

airindia-delay-predictor/
├── data/
│   ├── raw/
│   ├── processed/
│   └── external/
├── notebooks/
│   ├── 01_data_collection.ipynb
│   ├── 02_feature_engineering.ipynb
│   ├── 03_model_training.ipynb
│   ├── 04_evaluation.ipynb
│   └── 05_shap_explainability.ipynb
├── src/
│   ├── data_ingestion.py
│   ├── feature_engineering.py
│   ├── train_classifier.py
│   ├── train_regressor.py
│   ├── evaluate.py
│   ├── predict.py
│   └── utils.py
├── frontend/
│   └── app.py
├── models/
│   ├── delay_classifier.pkl
│   └── delay_regressor.pkl
├── requirements.txt
└── README.md

Installation

1. Clone or download the project.
2. Install dependencies:

pip install -r requirements.txt

Data Preparation

1. Run data ingestion:

python src/data_ingestion.py

This generates sample flight data. In production, replace with real data sources.

2. Run feature engineering:

python src/feature_engineering.py

Model Training

1. Train classifier:

python src/train_classifier.py

2. Train regressor:

python src/train_regressor.py

Evaluation

Run evaluation:

python src/evaluate.py

Running Notebooks

Open the notebooks in Jupyter:

jupyter notebook notebooks/

The notebooks provide detailed walkthroughs of each step.

Launch Frontend

1. Start the API server:

python api.py

2. In a new terminal, start the Astro frontend:

cd frontend
npm install
npm run dev

3. Open http://localhost:4321 in your browser

The API runs on http://localhost:8000 and serves predictions to the Astro frontend.

Enter flight details and get predictions.

Retraining Strategy

• Update rolling statistics daily
• Full model retrain monthly
• Models versioned with timestamps

Data Sources

Primary Data Sources (Real)

1. Actual Flight Times (ADS-B): OpenSky Network

   • Source: https://opensky-network.org/data/impala
   • Data: Actual departure/arrival timestamps, callsigns (AIC for Air India)
   • Access: Free signup required, academic/research use allowed
   • Tables: flights_data4, flightlist, states
   • Example Query:

     SELECT callsign, origin, destination, firstseen, lastseen
     FROM flights_data4
     WHERE callsign LIKE 'AIC%' AND year = 2023;
     

2. Flight Schedules: Air India Timetables or OpenFlights

   • Source: Air India website or https://openflights.org/data.html
   • Data: Scheduled departure/arrival, routes, distances
   • Files: routes.dat, airports.dat

3. Airport Delay Statistics: DGCA (India)

   • Source: https://www.dgca.gov.in/digigov-portal/?page=4264/4206/sericename
   • Data: Monthly on-time performance, airline/airport delay stats
   • Use: Validation, congestion priors, baseline rates

Current Implementation

• Simulated data for demonstration
• Real airport metadata included
• Weather: Simulated (Meteostat ready)
• Replace with real sources for production

Setup for Real Data

1. OpenSky Network (Flight Times)

1. Sign up at https://opensky-network.org/data/impala
2. Get Impala credentials
3. Install Impala client: pip install impyla
4. Update data_ingestion.py with credentials

2. Flight Schedules

• OpenFlights data is automatically downloaded
• For Air India specific schedules, scrape from airline website

3. DGCA Statistics

• Download monthly reports from DGCA website
• Parse PDFs or CSVs for delay statistics

4. Weather (Meteostat)

• Sign up at https://meteostat.net/
• Update weather fetching code with API key if needed

Running with Real Data

# Fetch real schedules
python -c "from src.data_ingestion import fetch_flight_schedules; print(fetch_flight_schedules())"

# For OpenSky: Update credentials and run custom query

Production Deployment

• Use real data APIs for flight schedules and weather
• Deploy models on cloud (AWS SageMaker, etc.)
• Schedule daily updates for features
• Monitor model performance

✅ FINAL GOAL

At the end, I should be able to:

pip install -r requirements.txt
python api.py &
cd frontend && npm install && npm run dev

And get accurate Air India delay predictions for any route in a modern web interface!