フリー問題

問題文

A regression model predicts the number of minutes a repair will take. Which metric should the monitor track over time to detect degradation?

選択肢

1. An error metric in the units of the target, such as the root mean squared error, computed once the actual durations are known.
2. The confusion matrix of the predicted and the actual durations, one row per repair, because it shows exactly where the model over-predicts and where it under-predicts.
3. The share of correct predictions, reported by the monitor once in each monthly window, computed by counting the predictions that matched the actual duration exactly.
4. The area under the curve, because it is bounded between zero and one whatever the predicted repair minutes turn out to be, and is therefore comparable across periods.