American Airlines
Forecasting where the bags go
- 100kmishandled bags a year, projected to be prevented
- 15%forecast accuracy gain
- 1stplace in the contract competition
I built a PyTorch time-series model on more than 3 million baggage check-in samples to forecast baggage flow. It’s projected to prevent 100,000 mishandled bags a year, with a 15% accuracy gain.
A custom loss combining Earth Mover’s Distance and RMSE raised forecast accuracy another 8%. I presented the predicted probability distributions to operations for logistics planning.