ML Fundamentals on a Fraud Model: Overfitting, Imbalance, Optimizers, Metrics, Transformers

MediumClassical ML

Reported in interviews at Snap

Problem statement

How would you improve a model that flags fraudulent online card payments? Fraud is rare, and flagged payments require extra verification. Missing fraud is costly, but needless verification also creates friction.

Explain overfitting, the cost-based threshold after sampling, and how to choose a ranking for a limited review queue.

Show hint

For each part, ask which failure mode the technique exists to fix, and what evidence in the numbers points to it.

Relevant roles: ML Engineer, Data Scientist and Applied Scientist.

Related free problems