Hotel Booking Cancellation Prediction

Consists of a full-stack MLOps pipeline for real-time predictions and automation. The model is uploaded as a container to heroku which serves real time predictions and is scheduled using Airflow DAGs.

Machine Learning & MLOps

  • Model Development: XGBoost classifier trained on real-world booking data, optimized for accuracy and fairness.

  • Data Pipeline: Automated DataOps & MLOps workflows using Apache Airflow for ingestion, transformation, and model retraining.

  • Monitoring & Governance: SHAP values for explainability, drift detection to ensure model relevance, and a feedback loop for continuous learning.


Flowchart


Impact

Increased revenue by proactively managing overbookings.

Enhanced resource planning with real-time insights.

Improved customer retention through smarter cancellation policies.


Deployment & Hosting

  • Containerization: Dockerized model packaged for portability and scalability.

  • CI/CD Pipeline: GitHub Actions automates testing and deployment.

  • Hosting: The FastAPI model API and Flask dashboard are deployed on Heroku, with separate staging and production environments.

  • Daily Predictions: Integrated with a MySQL database and scheduled via Airflow DAGs, triggering batch predictions and logging insights for hotel managers.