99% and Still Wrong

A fraud detection model flags transactions as fraudulent or legitimate. Only 1% of transactions are actually fraudulent. A model that predicts 'legitimate' for every single transaction achieves 99% accuracy. What does this scenario best illustrate?

Precision and recall are undefined whenever classes are imbalanced

Accuracy is misleading on imbalanced datasets because a trivial majority-class classifier can score high while catching zero positives

Accuracy is always the best metric regardless of class balance

The model must be overfitting to the training data