Design and sample for credit default prediction

A card-issuing bank wants a model that flags accounts likely to default on a credit card within a 90-day horizon, scored at account-month granularity, so that a retention team can reach out before the loss lands. Defaults are rare in production: roughly 2% of account-months end in default. You are asked to lay out the complete end-to-end design, with particular depth on how the training sample is constructed. 1. Label, features and split. Define the label exactly — the…