Build a churn prediction model.
Use only information available at prediction time and predict a future outcome.
Which data preparation is correct?
Use prediction-time customer attributes as features and future churn as the label, while recording timing and meaning.
Detailed explanation
That value is unavailable at prediction time and causes leakage.
Leakage and reproducibility could not be checked.
This prevents future information from entering the inputs.
The model would receive the answer directly.
Check official documentation
Compare the following points with the official documentation.
Review data preparation in Task 1.3 and the AWS guidance on unavailable production features. Choose the churn-prediction time and compare when customer attributes, post-churn refunds, and the outcome label become available.
Official references
What to confirm
Separate input features from the target and explain why information unavailable at prediction time must be excluded from the inputs.
Key points
- Features
- Labels
- Observation time
Notes
- Environment: Reference review of official documentation and hypothetical examples; no AWS resources or model training are required.
- An outcome label may become available later. Features for historical training examples must still reflect what was available at the original prediction time.
- Record the churn horizon and feature-availability times. A collection timestamp alone may not show when information became usable.
Foundation review
Availability boundary
A feature is valid only when the production process can obtain it before the prediction is made.
Target separation
The future outcome must remain separate from input features used to predict it.