Only 0.5% of card payments are fraud. A model that always predicts "not fraud" has what accuracy?
Options
- 99.5% (Correct option)
- 50%
- 100%
- 0.5%
Answer
99.5%
Why
Accuracy is the share of correct predictions. If 995 of every 1,000 payments are legitimate, always saying "not fraud" is right 99.5% of the time.
Yet this model catches no fraud at all. With imbalanced classes, accuracy hides the failure.
Look at precision, recall or PR-AUC for the rare class instead.
Remember
Accuracy misleads when one class is rare.