Flight Price Prediction

A machine learning system for predicting flight ticket prices using Python, scikit-learn, XGBoost, CatBoost, and FastAPI.

Problem Description

Airlines dynamically price tickets based on multiple factors including:

• Route: Source and destination cities affect base pricing
• Timing: Departure time, day of week, and season impact prices
• Airline carrier: Different airlines have different pricing strategies
• Number of stops: Direct flights typically cost more than connecting flights
• Booking class: Business vs economy affects pricing significantly

This system predicts flight prices to help:

• Travelers: Find optimal booking times and routes
• Airlines: Optimize pricing strategies
• Travel agencies: Provide better price estimates

The model is trained on the Kaggle Flight Price Prediction Dataset and achieves R² > 0.85 using random forest algorithm.

Features

• Comprehensive EDA with visualizations
• 6 ML models trained and compared
• Hyperparameter tuning with RandomizedSearchCV
• FastAPI REST API for predictions
• Docker containerization for deployment
• Full reproducibility with random_state=42

Screenshots

Exploratory Data Analysis

Dataset Overview	Missing Values

Price Distribution	Feature Distributions

Correlation Heatmap	Datetime Features

Feature Importance

Model Training

Model Comparison	Best Model Performance

Hyperparameter Tuning

API Documentation

Swagger UI	ReDoc

API Testing

Dataset

Download the dataset from Kaggle and place it in the dataset/ folder:

1. Go to Kaggle Flight Price Prediction Dataset
2. Download the dataset (requires Kaggle account)
3. Extract and place the CSV file:

mkdir -p dataset
mv ~/Downloads/Clean_Dataset.csv dataset/

The dataset directory is gitignored due to file size.

Environment Setup

Prerequisites

• Python 3.13+
• uv package manager

Installation

# Install uv (if not already installed)
curl -LsSf https://astral.sh/uv/install.sh | sh

# Clone repository
git clone https://github.com/devnovikov/flight-price-prediciton.git
cd flight-price-prediciton

# Create virtual environment and install dependencies
uv sync

# Activate virtual environment
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

Training

Option 1: Jupyter Notebook (Recommended for EDA)

# Start Jupyter
uv run jupyter notebook notebooks/notebook.ipynb

The notebook includes:

• Complete exploratory data analysis
• Feature engineering
• Model training and comparison
• Hyperparameter tuning
• All visualizations saved to screenshots/

Option 2: Production Training Script

# Train model using production script
uv run python src/train/train.py

Expected output:

============================================================
FLIGHT PRICE PREDICTION - MODEL TRAINING
============================================================

Random State: 42
Loading data from dataset/Clean_Dataset.csv...
  Loaded 300,153 rows, 12 columns
Preprocessing data...
  Features shape: (300153, 11)

============================================================
TRAINING MODELS
============================================================

Baseline models:
  Training Linear Regression...
  Training Ridge Regression...
  Training Lasso Regression...

Models with hyperparameter tuning:
  Tuning Random Forest...
  Tuning XGBoost...
  Tuning CatBoost...

============================================================
MODEL COMPARISON (sorted by RMSE)
============================================================
            Model        RMSE       R2         MAE
    Random Forest 2127.354825 0.991221  768.780202
          XGBoost 2189.536252 0.990700 1069.969849
         CatBoost 3084.850592 0.981539 1718.064701
 Lasso Regression 7012.694003 0.904598 4622.719640
 Ridge Regression 7012.704891 0.904598 4623.039426
Linear Regression 7012.705115 0.904598 4622.989854

============================================================
BEST MODEL
============================================================

Best Model: Random Forest
  RMSE: 2,127.35
  R2: 0.9912

Best Hyperparameters: {'n_estimators': 300, 'min_samples_split': 2, 'min_samples_leaf': 1, 'max_depth': None}

API Usage

Start the Server

# Start FastAPI server
uv run uvicorn src.predict.predict:app --reload

Server will start at http://localhost:8000

Interactive Documentation

• Swagger UI: http://localhost:8000/docs
• ReDoc: http://localhost:8000/redoc

API Endpoints

Health Check

curl http://localhost:8000/

Response:

{
  "status": "healthy",
  "model_loaded": true,
  "version": "1.0.0"
}

Predict Price

curl -X POST http://localhost:8000/predict \
  -H "Content-Type: application/json" \
  -d '{
    "airline": "IndiGo",
    "source": "Delhi",
    "destination": "Cochin",
    "date_of_journey": "2026-03-15",
    "dep_time": "10:30",
    "arrival_time": "14:45",
    "duration": "4h 15m",
    "total_stops": "1 stop"
  }'

Response:

{
  "predicted_price": 8542.75,
  "currency": "INR",
  "model_version": "1.0.0",
  "features_used": ["Journey_day", "Journey_month", "..."]
}

Docker

Build Image

# Ensure model is trained first
uv run python src/train/train.py

# Build Docker image
docker build -t flight-price-api .

Run Container

# Run container
docker run -p 8000:8000 flight-price-api

Test Container

# Health check
curl http://localhost:8000/

# Make prediction
curl -X POST http://localhost:8000/predict \
  -H "Content-Type: application/json" \
  -d '{
    "airline": "IndiGo",
    "source": "Delhi",
    "destination": "Cochin",
    "date_of_journey": "2026-03-15",
    "dep_time": "10:30",
    "arrival_time": "14:45",
    "duration": "4h 15m",
    "total_stops": "1 stop"
  }'

Project Structure

flight-price-prediciton/
├── dataset/                    # Kaggle dataset (gitignored)
│   └── Clean_Dataset.csv
├── models/                     # Trained model artifacts
│   └── best_model.joblib
├── notebooks/
│   └── notebook.ipynb          # EDA and model training
├── screenshots/                # Visualizations (13 images)
├── src/
│   ├── __init__.py
│   ├── train/
│   │   ├── __init__.py
│   │   ├── constants.py        # Configuration constants
│   │   ├── preprocessing.py    # Data preprocessing
│   │   └── train.py            # Production training script
│   └── predict/
│       ├── __init__.py
│       ├── schemas.py          # Pydantic models
│       ├── model_loader.py     # Model loading utility
│       ├── preprocessing.py    # Inference preprocessing
│       └── predict.py          # FastAPI application
├── tests/
│   └── __init__.py
├── specs/                      # Feature specifications
├── .gitignore
├── .dockerignore
├── pyproject.toml              # Dependencies (uv)
├── uv.lock                     # Locked versions
├── Dockerfile
└── README.md

Models Trained

Model	RMSE	R² Score	MAE
Random Forest	2127.354825	0.991221	768.780202
XGBoost	2189.536252	0.990700	1069.969849
CatBoost	3084.850592	0.981539	1718.064701
Lasso Regression	7012.694003	0.904598	4622.719640
Ridge Regression	7012.704891	0.904598	4623.039426
Linear Regression	7012.705115	0.904598	4622.989854

Best model (Random Forest) achieves R² > 0.85 target.

Reproducibility

• Random State: All random operations use random_state=42
• Dependencies: Locked with uv.lock for exact versions
• Dataset: Download instructions provided above

Cloud deployment

Deploy Docker with CLI on Google Cloud

The current model can be deployed to Google Cloud for prediction. Follow the steps.

1. Install prerequisites

# Install Google Cloud SDK
curl https://sdk.cloud.google.com | bash
exec -l $SHELL

# Login and set project
gcloud auth login
gcloud config set project YOUR_PROJECT_ID

2. Enable required APIs

gcloud services enable run.googleapis.com cloudbuild.googleapis.com

3. Build and push to Google Container Registry

gcloud builds submit --tag gcr.io/YOUR_PROJECT_ID/flight-price-api

4. Deploy to Cloud Run

gcloud run deploy flight-price-api \
  --image gcr.io/YOUR_PROJECT_ID/flight-price-api \
  --platform managed \
  --region us-central1 \
  --memory 4Gi \
  --cpu 2 \
  --timeout 600 \
  --allow-unauthenticated

Tech Stack

• Language: Python 3.13
• Package Manager: uv
• ML Frameworks: scikit-learn, XGBoost, CatBoost
• API: FastAPI, uvicorn, Pydantic
• Visualization: matplotlib, seaborn
• Containerization: Docker

License

This project is for educational purposes. The dataset is from Kaggle and subject to its license terms.