Aviation Shipping and Transportation Analytics

GitHub: https://github.com/gw-00/ADS-508-Project

Predicting Flight Delays with AWS & Machine Learning

This project focused on helping a logistics company reduce cargo delays at San Diego International Airport. I built a full machine learning pipeline using AWS SageMaker, S3, and Athena to clean and merge weather + flight data. Using XGBoost, I trained a classification model that reached 95% recall—accurately predicting which flights would be delayed or canceled. The goal was to support better planning and reduce unplanned disruptions for air cargo shipment.

Full Paper Report:

Final Project Report-Group9.pdf

Download PDF • 2.54MB

Final Presentation for Non Technical Audience

https://youtu.be/P7ZnX8hrD-8

Results & Impact

• Achieved 95% recall on the delayed/canceled flight class, ensuring the model caught nearly all disruptions.

• Used AWS SageMaker's built-in XGBoost algorithm for scalable, cloud-based training and deployment.

• Reduced class imbalance through undersampling, leading to more accurate predictions across flight statuses.

• Connected real-time weather patterns to flight outcomes, helping identify key contributors like wind speed and visibility.

• Built a modular pipeline with S3 buckets, Athena queries, and a clear data flow from raw ingestion to predictions.

• Designed the model as a decision-support tool, helping logistics teams plan around potential delays more effectively.