S05 · adr simulation

ADR: one global demand model or one model per route

People, data and outputs in this scenario are simulated. Each stage is recorded when you commit; the reveal comes after.

4stages
35 minabout
0stages committed

The brief

The synthetic airline's planning team (simulated) wants a daily forecast of passengers per flight, seven days ahead. The network has 11 routes and 19 flights today and plans several new routes a year. The team is two data scientists and one ML engineer. The model would feed weekly capacity and rostering decisions.

Evaluation figures below come from the course's synthetic flight table: models trained on 2022-2023, evaluated on 2024 (6,954 flight-days). Features: weekday, holiday flag, month, season, event flag, seats, the same weekday last week and the mean of the last four same weekdays. All people and constraints are simulated.

Stage 1 of 4

A colleague (simulated) says: "Routes differ a lot, so obviously each route needs its own model." Before comparing options you list the facts that will decide.

Which facts must you establish before choosing? Select all that apply.

How often new routes launch and what the forecast must do for a route that has little or no history.What a simple seasonal baseline achieves on the same evaluation, since that is what any model must beat.How many models the team can monitor, retrain and roll back with one ML engineer.Which model type tops public forecasting leaderboards, so that the choice rests on external evidence.Whether all routes have the same number of daily flights, since a global model needs equal volumes.