The Classifier BoundaryThreshold: 0.50
DECISION LINE
Safe Region
Alarm Region
88%
Precision
88%
Recall
Confusion Matrix
Actual \ Pred
Predicted Legit
Predicted Fraud
Total Legit
44
True Neg
6
False Pos
Total Fraud
6
False Neg
44
True Pos
The Trade-off
BALANCED: You are trading off customer friction vs insurance risk.
Confusion Matrix
Precision measures "how often are we right when we say YES?", while Recall measures "did we catch all the YESes that existed?". You almost always have to sacrifice one to get the other.
Founder Strategy
In Fraud Detection, you want high Recall (catch every thief). In Medical AI, you want high Precision (don't give surgery to healthy people). Understand your product's "Cost of a Mistake" before picking your threshold.