Payment fraud detection models work best as a layered system: interpretable supervised models for known fraud, anomaly detectors for new attacks, and graph models for connected account abuse. For payment companies, the practical goal is…