Practice / Data & AI

Spot the data leakage

Medium · Whiteboard

A teammate's churn model has 0.98 AUC. Features include: tenure, plan type, support tickets in the last 30 days, "account_closed_reason", and number of logins in the 30 days after the prediction date. The train/test split is random. What is wrong?

Create a free account to answer it and get scored on what a strong answer covers.