S18 · lifecycle simulation
A route forecast from intake to handover
People, data and outputs in this scenario are simulated. Each stage is recorded when you commit; the reveal comes after.
- 5stages
- 50 minabout
- 0stages committed
The brief
Network planning at the synthetic airline wants a daily forecast of passengers per flight for 19 flights on 11 routes
to plan capacity and fare actions up to 90 days ahead. The data are the course database (flight_day_raw for the
2022-2023 feed, flight_day for the curated series, 2022-2024). The model is a gradient-boosting regressor on features
known in advance (route, flight, weekday, month, day of year, holiday, event flag, capacity, seasonality label and a
trend term). The baseline is "same flight, same weekday last year" (364 days back). The error metric is WAPE (sum of
absolute errors over sum of actual passengers) with signed bias (forecast minus actual over actual).
Dollar figures, compute costs and staffing in this scenario are hypothetical. The course database also holds a
competitor_capacity column; nothing in the first model uses it.
Stage 1 of 5
Audit of flight_day_raw (13,887 rows against 13,870 in the curated 2022-2023 series):
| Finding | Count |
|---|---|
| Duplicate flight-day keys (some with a late revenue update) | 110 |
| Route codes in lower case | 124 |
| Rows with load factor above 1 | 6 |
| Rows with negative bookings | 2 |
| One passenger count ten times too large (840) | 1 |
Rows missing avg_fare and revenue | 281 |
| IST-TZX rows missing for 2023-03-04 to 2023-03-12 | 9 days |
Which findings must be dealt with before training a model for passengers? Select all that apply.