Design a hybrid marketplace fraud system

Build a fraud detection system for (a) fake marketplace listings and (b) fabricated education credentials. 1) List at least 10 features spanning unique identifiers (device ID, IP, payment instruments), text/image signals (low‑res, duplicate embeddings), behavioral patterns (burst account creation, connection request acceptance rates), and network features (triadic closure among 'alumni'). 2) Combine supervised learning (to catch known patterns) with anomaly detection (to flag novel attacks); specify algorithms and how to resolve class imbalance and noisy labels. 3) Choose an operating threshold using explicit costs of false negatives vs. false positives and show how that maps onto ROC/PR curves. 4) Propose a risk‑based step‑up authentication policy (e.g., <10% auto‑pass, 10–50% 2FA, >50% auto‑block) and an online evaluation plan that limits collateral damage to legitimate users. 5) Define concept‑drift monitoring, retraining cadence, and red‑team simulation to uncover adaptive fraud.