A churn model uses a field recorded only after a customer has churned.
The offline score is high, but the field is unavailable at prediction time.
Which mitigation is appropriate?
Use only data available at prediction time and design time-aware splits so future information cannot enter training.
Detailed explanation
This prevents future information from making the offline score unrealistically high.
Creating unavailable information is not a valid production design.
This removes evaluation independence and worsens leakage.
The high score may be caused by leakage.
Check official documentation
Compare the following points with the official documentation.
AWS公式AIF-C01ガイドDomain 1.3のデータ品質と評価設計を確認する。
Official references
What to confirm
予測時点、将来情報、時間順分割の関係を説明できる。
Key points
- Future information
- Prediction time
- Leakage
Notes
- Environment: AWS公式AIF-C01試験ガイドとAWS公式ドキュメントの確認
- Compare the definition with the AIF-C01 objective and the service documentation before applying it.
- Keep the concept separate from a particular answer choice or implementation step.
Foundation review
Causal order
The training process must respect the order in which information becomes available in the real workflow.
Evaluation integrity
Leakage invalidates a comparison because the model receives information it would not have in production.