Prediction accuracy
Shrinkage reduction
Months development
Since 2016, the airline has struggled significantly with inaccurate carry-on bag predictions, achieving less than 20% reliability in their estimates. This consistently led to inefficient baggage handling processes and frequent flight delays.
Key challenges
Absence of real-time bin space monitoring and feedback mechanisms
Lack of systematic data integration across data sources and systems
Low adoption due to unexplainable predictions
Inflexible batch deployment system
The solution
Intelligent data processing
CatBoost Regression decision tree ensemble
Real-time API-driven model updates
Comprehensive feature engineering
Enterprise integration
AWS-based flexible architecture
User-friendly front-end application
Continuous model retraining capability
1
Foundation
Created flexible architecture
Developed prediction requirements
Integrated diverse data sources
2
Development
Built machine learning models
Applied sampling techniques
Refined from user feedback
3
Deployment
6-9 month design and development phase
Continuous improvement process
3-month implementation
Expanding predictive capabilities
Further improving operational streamlining
Enhanced intelligence
Process optimization
Identifying critical high-risk categories
Risk management
