問題文
A loan approval model's outcome is only known 60 days after the prediction. The team wants to track accuracy over time in the same profile. How should they handle this delay?
選択肢
- Write the outcome back into the same request log rows when it arrives, and let the refresh recompute the metrics for the affected windows.
- Use predicted scores as a proxy for the outcome until the real outcome arrives, then replace the metric values, so a trend line exists meanwhile.
- Create a second profile on a separate table of outcomes and compare its metrics to the prediction profile by hand, joining the two sets of metrics in a spreadsheet.
- Delay the profile refresh by 60 days so that every window already has its outcomes when it is first computed with outcomes attached.