For KLM, operating a flight begins years before take off. The Dutch airline must first decide where to fly, when to fly, the crews that are needed, etc. Passenger forecasts are needed to help make these decisions.
The company needs to predict not only how many people will purchase flight tickets, but also how many will actually board the plane, which type of seat they will have, and how much baggage. There are also many different potential users for this model, all with various criteria.
In this webinar, Meeke, a former Xomnia junior data scientist, explains why the team chose a Multi Task neural network for the model. She details the pros and cons of multi and single-task neural networks for machine learning.