Airline Performance Dashboard

An interactive Streamlit dashboard for analyzing US domestic flight delays with machine learning predictions.

Features

• Delay Analysis: Comprehensive breakdown of delay types and causes
• Time Analysis: Temporal patterns in flight volumes and delays
• Airline Analysis: Performance metrics by carrier
• Airport Analysis: On-time performance by airport
• Deep Dive (EDA): Exploratory data analysis with detailed insights
• ML Prediction: Interactive flight delay probability prediction
• About: Complete data science lifecycle documentation

Installation

1. Clone the repository or download the files
2. Install dependencies:

pip install -r requirements.txt

Data Setup

Data Source

The dataset contains 5.8 million US domestic flight records from 2015, sourced from the US Department of Transportation via Kaggle.

Data Format

The dashboard uses optimized Parquet format for fast loading:

• Original: ~500MB CSV file
• Optimized: 74MB Parquet file (85% reduction)

If you have the original flights.csv, it will be loaded automatically. For better performance, convert it to Parquet format using the included script.

Machine Learning Model

Option 1: Pre-train the Model (Recommended)

For faster dashboard loading, pre-train the model once:

python train_model.py

This will:

• Train a Random Forest classifier on the flight data
• Evaluate the model and display metrics
• Save the trained model to flight_delay_model.pkl

The dashboard will automatically load this pre-trained model for instant predictions.

Option 2: Train On-the-Fly

If you don't pre-train the model, the dashboard will train it automatically on first load of the ML Prediction tab. This takes about 1-2 minutes but is cached for the session.

Running the Dashboard

streamlit run app.py

The dashboard will open in your browser at http://localhost:8501

Project Structure

├── app.py                      # Main Streamlit dashboard
├── utils.py                    # Data loading and ML utilities
├── train_model.py              # Standalone model training script
├── flights.parquet             # Optimized flight data
├── airlines.csv                # Airline reference data
├── airports.csv                # Airport reference data
├── requirements.txt            # Python dependencies
└── README.md                   # This file

Data Science Lifecycle

This project demonstrates a complete data science workflow:

1. Problem Definition: Predict flight delay probability
2. Data Collection: 5.8M flight records from US DOT via Kaggle
3. Data Preprocessing: CSV to Parquet optimization, data cleaning
4. EDA: Interactive visualizations across multiple dimensions
5. Modeling: Random Forest classifier with 8 features
6. Evaluation: Accuracy, Precision, Recall, F1-Score
7. Deployment: Interactive Streamlit dashboard

Model Details

• Algorithm: Random Forest Classifier
• Features: Airline, Origin/Destination Airports, Month, Day of Week, Day, Scheduled Departure, Distance
• Target: Binary classification (Delayed >15 min vs On-Time)
• Training: 80/20 train-test split with stratification
• Performance: See ML Prediction tab for live metrics

Technical Optimizations

• Parquet Format: 85% size reduction with gzip compression
• Categorical Dtypes: Efficient memory usage for string columns
• Model Caching: Train once, use across sessions
• Pre-computed Metrics: Dashboard metrics calculated once and reused
• Parallel Processing: Multi-core Random Forest training

Requirements

• Python 3.8+
• streamlit
• plotly
• pandas
• scikit-learn

License

This project is for educational purposes demonstrating data science best practices.