A team building a spam-email filter reviews their model's mistakes and decides that a legitimate email being wrongly marked as spam and hidden from the user is far more damaging than an actual spam email slipping into the inbox. Which evaluation metric should the team prioritize maximizing when tuning the model?
Select an answer to reveal the explanation.
Short Explanation
The team has told us exactly what kind of mistake worries them most: a good email getting caught in the spam trap and never seen. That is a false-alarm problem, so the metric to lean on is the one that answers the question, of everything the filter called spam, how much of it was actually spam. Pushing that number up means the filter gets pickier about what it labels spam, which cuts down on innocent emails getting swept up by mistake. Chasing the opposite goal - catching every last piece of spam no matter what - would actually make the false-alarm problem worse, since a filter trying hard not to miss any spam tends to get trigger-happy and snags more legitimate mail along the way. The other two candidates do not even apply to a yes-or-no decision like this one; they are built for problems where the model is predicting a number on a scale, like a price or a temperature, not sorting messages into two buckets. Given the team's stated worry, the pickiness-focused metric is the right one to chase.
Full Explanation
The correct answer is A. Precision measures, out of everything the model labeled as positive (in this case, spam), how many of those labels were actually correct, so maximizing precision directly reduces the number of legitimate emails incorrectly flagged as spam, which is the exact cost the team wants to minimize. Option B is incorrect because recall measures how many of the actual spam emails the model successfully caught, and prioritizing it would push the model toward flagging more emails as spam overall, increasing rather than decreasing the risk of legitimate email being misclassified. Option C is incorrect because mean absolute error is a metric used to evaluate regression models that predict continuous numeric values, such as predicted prices or temperatures, and does not apply to a classification decision like spam versus not spam. Option D is incorrect because R-squared measures how well a regression model's predictions explain the variance in a continuous outcome, which again is a regression concept and has no meaning for a binary spam classification task. Since the team's stated concern is specifically about false positives, that is, legitimate email wrongly caught by the filter, the metric that directly penalizes that error is the one they should prioritize.