Question 300
A production drift monitor produces noisy alerts for dozens of unimportant text-like feature columns. The team is preparing a controlled change and wants evidence that it corrects the root cause. After remediation, the operator will verify that alerts concentrate on the approved relevant features under controlled test drift. What is the best next action for this Azure AI workload?
Explanation
Correct Answer
Monitoring every low-value field can obscure operationally meaningful distribution shifts. On retest, alerts concentrate on the approved relevant features under controlled test drift.
Incorrect Answers
Numerical and categorical features need suitable distribution-comparison statistics. By contrast, monitoring every low-value field can obscure operationally meaningful distribution shifts.
Detection cadence must follow the speed of production input distribution change; monitoring every low-value field can obscure operationally meaningful distribution shifts.
That choice is useful when an ML monitor shows drift for almost every column because it compares production against an unrelated test dataset; monitoring every low-value field can obscure operationally meaningful distribution shifts.