AI Fraud Detection Quiz

Answers and explanations appear as you take the quiz.

1. Why are fixed rules alone (e.g., 'block charges over $5,000') often inadequate for fraud?
   • Fraud is adversarial and constantly changing, so rules go stale
   • They are too expensive to write
   • Banks are not allowed to use rules
   • Rules only work on weekends
2. 'Class imbalance' in fraud detection refers to the fact that:
   • There are too many banks
   • Fraudulent transactions are a tiny fraction of all transactions
   • All customers spend the same amount
   • Data is stored in the wrong class file
3. A 'false positive' in fraud detection means:
   • A real fraud was missed
   • The system crashed
   • A legitimate transaction was wrongly flagged as fraud
   • A duplicate charge occurred
4. What challenge can neighbor-based message propagation create for a fraud GNN?
   • It guarantees every linked account is fraudulent
   • It removes all concept drift from the data
   • It prevents investigators from reviewing transactions
   • Neighbor and hub effects can amplify class-imbalance problems
5. Which signal is an example of 'velocity' in transaction monitoring?
   • Many charges occurring within a few minutes
   • The customer's birthday
   • The color of the card
   • The bank's logo
6. What fraud typology did FinCEN’s 2024 deepfake alert describe?
   • Deepfakes that automatically reverse settled transactions
   • Deepfake media and fraudulent identity documents used to evade identity verification or authentication
   • A rule requiring banks to reject every voice call
   • Synthetic identities that prove a customer committed a crime
7. Why can accuracy be misleading for a highly imbalanced fraud dataset?
   • Accuracy is always equal to recall
   • Fraud labels are never used
   • A model can score high accuracy by predicting the majority legitimate class while missing fraud
   • The test set must contain no legitimate transactions
8. What is one reason a graph model can help identify linked fraud patterns?
   • It removes the need for labels or investigation
   • It proves that connected accounts are criminal
   • It guarantees real-time detection
   • It can represent relationships among accounts, transactions, or devices