Taxi Demand Forecasting at Changi Airport

Project Context

Developed during my internship at Changi Airport Group to optimize taxi allocation and reduce passenger wait times.

Challenge

Accurately predict taxi passenger demand across different terminals and time periods to improve operational efficiency.

Solution

Model Development

• Implemented XGBoost regression model
• Engineered features from historical data:
  • Flight schedules and turnaround times
  • Seasonal patterns
  • Day of week and time of day
  • Weather conditions
  • Special events

Performance Improvement

Increased forecasting accuracy from 48% to 62% through:

• Feature engineering and selection
• Hyperparameter tuning
• Ensemble methods

Deployment

• Built CI/CD pipeline for automated deployment
• Utilized AWS services:
  • CloudFormation for infrastructure as code
  • ECR for container registry
  • Lambda for serverless inference
• Implemented monitoring and alerting

Data Pipeline

• Conducted EDA on 18,000+ flight turnarounds
• Identified operational bottlenecks
• Built automated data pipeline on Amazon SageMaker

Business Impact

The improved forecasting enables better resource allocation, reducing passenger wait times and improving overall airport experience.