Whenever you're ready, walk me through how you'd approach this.
How would you estimate the probability that a user is fraudulent given transaction data?
Explain how you would define the unit of analysis, construct fraud labels, select transaction-level features, train a probabilistic model, and calibrate its output. Distinguish the probability of a fraudulent transaction from the probability that the user is fraudulent, and describe how you would evaluate performance when fraud is rare. Do not assume that an unlabeled transaction table alone identifies fraud.