Forecasting Airline Passengers Using Designer Machine Learning
summary
Forecasts passenger counts for individual flights using a progression of models from a simple linear regression to feed-forward and recurrent deep neural networks (Keras examples shown). Covers feature engineering from artificial flight-booking data, loss choices (MSE, conditional density, mixture density networks estimating mu/sigma), sequence feature extraction with RNN/LSTM, system design constraints (shrinking horizon), and evaluation for use cases like aircraft allocation and probability of capacity overflow.
AI-written summary · about
chapters (31)
- 0:00Introduction
- 0:50Problem: Predicting Passenger Number & Use Cases
- 1:38Problem: Unique Forecasting Constraint - Shrinking Horizon
- 2:08System Requirements
- 2:50System Design
- 3:23"Designer Machine Learning" Definition
- 3:56Data: Artificial Flight-bookings
- 5:07Data: Features
- 6:24Model: Simple Linear Model & ANN
- 7:15Model: Feed-Forward Deep Neural Network
- 7:48Model: Loss Function - MSE
- 8:25Keras Code Example
- 9:33Use Case: Aircraft Allocation
- 10:25Evaluation: Probability of Capacity Overflow
- 11:45Model: Conditional Density Estimation
- 13:48Model: Updated ANN Outputs (Mu & Sigma) & Loss Function
- 15:04Keras Code Example for Conditional Density Estimation
- 16:46Model: Mixture Density
- 17:09Model: Mixture Density Networks
- 17:51Challenges: Selecting Distributions & Numerical Optimization
- 19:06Sequence Feature Extraction
- 19:41Model: Representational Learning & Recurrent Neural Network
- 21:08Keras Code Example for RNN with LSTM
- 21:57Challenges: Non-uniform Time Deltas & Flight Dependencies
- 23:08Key Take-aways
- 24:32Q&A: Q1
- 26:04Q&A: Q2
- 27:26Q&A: Q3
- 29:16Q&A: Q4
- 30:58Q&A: Q5
- 32:40Q&A: Q6
related talks
32mBayesian Deep Learning and Flight Delay Prediction
37mForecasting using machine learning methods
34mMachine Learning to know if you R coming
24mPredicting Hotel Cancellations with Machine Learning
39mForecasting With Classical and Machine Learning Methods
40mMachine Learning
22mBefore the Model: How Machine Learning Products Start...
56mAutomatic Revenue Forecasting with Boosting Models and DL