Question: A classifier used in a hospital screening workflow treats every mistake as equally bad. In reality, missing a patient who needs urgent follow-up is far more serious than sending a low-risk patient for an extra review. Explain how a practitioner could change the evaluation metric to reflect that difference, and describe how the change would affect model comparison.

Sample answer: A practitioner can replace the plain error count with a cost-sensitive metric that assigns a much larger penalty to missed urgent cases. For example, a false negative that fails to identify a high-risk patient could count many times more than a false positive that sends someone for an unnecessary review. This makes the evaluation score match the real objective of the project: avoiding the most harmful mistakes. As a result, models that look similar under accuracy may rank very differently under the new metric, and the selected model will be the one that better protects patient safety.

Key points:

• Replace an equal-error metric with a weighted or cost-sensitive version.
• Give the most serious mistake a much larger penalty than less serious mistakes.
• Use a concrete high-stakes example such as missing a patient who needs urgent care.
• Let the new metric guide selection toward models that avoid the dangerous error, even if they make more minor mistakes.

Rubric: The response must explain how the metric is changed, why the change is needed to reflect the true cost of errors, and how the revised score changes which model is preferred during selection.