P80Feature EngineeringIntermediateDiagnose14 min
Diagnose Leakage From a Customer Lifetime Aggregate
Diagnose why a full-lifetime customer aggregate leaks future behavior into an earlier prediction.
#diagnose#leakage#aggregates
Scenario
A model predicts whether a customer will renew at day 30, but uses total_lifetime_spend calculated from the customer's entire account history.
Problem statement
Explain the leakage, why validation appears excellent, and how to rebuild the feature.
Your Task
- Identify information recorded after the prediction point.
- Contrast historical and full-lifetime aggregates.
- Define a valid replacement feature.
Related Concepts
Temporal leakageFeature availabilityPoint-in-time joins
Ready to count this problem?
Mark it complete when you have reasoned through the solution in your own words.