Bayesian Deep Learning and Flight Delay Prediction
summary
Freebird builds a real-time predictive engine that predicts both probability and magnitude of flight delays using a massive dataset — 1 billion vials of pricing and status streams — and finds an RNN outperforms random forests and linear regression for day-of and 24‑hour forecasts. The system uses Bayesian dropout (variational inference interpretation of dropout) to produce well‑calibrated uncertainty bands across airports; the model is live for internal rebooking agents and shows calibrated confidence that supports preemptive rebooking decisions.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:40Freebird
- 3:06Pricing
- 3:42Risk Model
- 5:49Database
- 6:45Data Access
- 9:33Challenges
- 11:46Bayesian Deep Learning
- 16:39Advantages of Bayesian Deep Learning
- 20:11Fixed Cost
- 20:51Three takeaways
- 21:54Team
- 23:03Questions
related talks
34mForecasting Airline Passengers Using Designer Machine Learning
42mFlying Python at 40,000 feet
34mMachine Learning to know if you R coming
39mBayesian Deep Learning with Edward (and a trick using Dropout)
26mReal Time Machine Learning
41mBayesian Network Modeling using R and Python
30mReal-Time ML services: From ideation to production
22mTaking Flight with AI and ML: Revolutionizing Airline Ops