Explain fraud types and evaluate a fraud model

You are interviewing for a Fraud Data Scientist role at PayPal. Answer the following: 1) List common fraud types relevant to payments (e.g., account takeover, first‑party fraud, third‑party fraud, merchant fraud). For each, give a short definition and an example. 2) Account Takeover (ATO): Explain how ATO typically happens end-to-end (attacker acquisition → credential takeover → monetization), and what signals/features you would expect to be useful for detection. 3) First‑party vs. third‑party fraud: - Define each clearly. - Explain why the label quality/ground truth can differ between them. - Describe at least one way mislabeling can bias model training/evaluation. 4) How would you measure the effectiveness of a fraud model in production? - Provide at least 5 metrics, including both ML metrics (e.g., precision/recall) and business/risk metrics (e.g., fraud dollars prevented, false-positive cost). - Describe the key tradeoffs (e.g., precision vs recall) and how decision thresholds should be chosen. - Clarify what the “positive class” is and how class imbalance affects evaluation. 5) You are asked to design a fraud strategy from scratch (“from 0”). Outline a practical plan across data, modeling, decisioning, monitoring, and iteration. Assume you must deploy something useful within 6–8 weeks. State any assumptions you need (e.g., time zone, label delay, chargeback window).