AI0-001 AI Concepts and Techniques Practice Question

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Reinforcement learning

  Why it's wrong here

  Reinforcement learning learns a policy from reward signals produced by interacting with an environment, not from a fixed set of labelled fraud outcomes. It is tempting because it also performs classification-like decisions, but it is correct for sequential decision problems such as adaptive transaction blocking, not static labelled prediction.

• ✗

  Unsupervised learning

  Why it's wrong here

  Unsupervised learning finds structure in data without labels, so it cannot map transactions to the known fraudulent or legitimate classes the scenario provides. It is tempting because fraud detection often uses clustering to surface anomalies, but that suits unlabelled data, whereas here historical labels already exist.

• ✓

  Supervised learning

  Why this is correct

  Supervised learning trains on labelled historical data, mapping inputs to known outputs. Fraud detection uses labelled transactions marked fraudulent or legitimate, so the model learns the mapping between transaction features and the fraud label, matching this paradigm exactly.

• ✗

  Self-supervised learning

  Why it's wrong here

  Self-supervised learning generates its own labels from unlabelled data via pretext tasks, so it cannot consume the historical fraud labels the scenario supplies. It is tempting because it also trains predictive models, but it is the right paradigm when labelled data is scarce and raw transaction records must be exploited instead.