Context

InboxShield uses a binary classifier to detect spam emails for a workplace email product. The team is deciding whether to keep the current threshold after complaints that too many spam emails still reach user inboxes, while some legitimate emails are incorrectly filtered.

Current Performance

Confusion Matrix Snapshot

On a validation set of 50,000 emails:

The Problem

The product manager points to 96.2% accuracy and argues the model is already strong. The support team disagrees because users still report missed spam, especially phishing-style messages. You need to explain what the F1 score means here and whether it is a better metric than accuracy for this use case.

Requirements

1. Define the F1 score and explain how it is calculated from precision and recall.
2. Interpret the current F1 score of 0.69 in the context of this spam detection problem.
3. Explain why accuracy may be misleading given the class imbalance.
4. Describe when F1 is more useful than accuracy, precision alone, or recall alone.
5. Recommend whether InboxShield should optimize for a higher F1 score or prioritize another metric.

Constraints

• False positives hide legitimate business emails and create user frustration.
• False negatives allow spam and phishing into inboxes.
• The team can adjust the classification threshold but cannot retrain the model this sprint.