29. A demand forecasting model looks accurate. What would you compare it against before recommending adoption?
MediumMachine learning basics
Technical
Show model answer
Use a time-based holdout and compare with credible simple baselines such as last period or the corresponding seasonal period. Choose evaluation horizons and error measures that reflect stockouts, waste and planning decisions.
Check performance across low-volume items, locations and peak periods, avoiding information from the future in features or tuning. Include operational cost and maintenance; a small average gain may not justify a complex model that fails on critical items.
What interviewers listen for
- Time-respecting evaluation
- Relevant simple baselines
- Segment and operational costs