Design anomaly detection and handle imbalanced logistic regression
Role: Data Scientist
Category: Machine Learning
Difficulty: medium
Interview Round: Technical Screen
Overview: Design a fraud detection setup for 50M transactions at a 0.4% fraud rate: a supervised logistic regression model plus an unsupervised anomaly detector for new merchants with no labels. Cover time-based splits, delayed labels, imbalance, calibration, and a cost-based threshold where a missed fraud costs far more than a false alarm.