P26Machine LearningAdvancedDiagnose18 min
Detect Data Leakage Before Deployment
Identify a feature that is highly predictive only because it is created after the outcome.
#leakage#time-splits#deployment#validation
Scenario
A churn model reaches validation AUC 0.97. One feature is cancellation_reason, recorded when an account closes.
Problem statement
Explain why this is leakage and how to redesign the feature set and validation split.
Your Task
- Define what information exists at prediction time.
- Remove or time-shift post-outcome features.
- Use a split that matches deployment chronology.
Related Concepts
Temporal validationFeature contractsDeployment realism
Ready to count this problem?
Mark it complete when you have reasoned through the solution in your own words.