Skip to content
ML Academy

Classical ML

Interview question

  • midlevel
  • 8 minsuggested time
  • 2Data scientist, ML engineer

Your insurance-claims fraud classifier has 99.2% accuracy and 0.93 ROC-AUC. Fraud is 0.8% of claims. The investigation team can review 200 claims a week. Is the model good? What would you report instead?

Tests: Metric choice under class imbalance and capacity constraints: precision at k, PR-AUC, recall at a fixed review budget, and cost-based evaluation.

day 13day 16

Sign in to practise: answers are saved before model answers are shown.