A fraud model includes a column containing the future outcome.
Validation is suspiciously high.
Choose two responses.
Use only information available at prediction time, split by time, and fit preprocessing on training data before applying it to validation.
Detailed explanation
Future information is not available during training.
Future or validation statistics can enter training.
The score may depend on impossible future information.
Validation information is less likely to leak into training.
Check official documentation
Compare the following points with the official documentation.
Review Task 1.3 and the leakage guidance. Identify columns finalized after the fraud-prediction time, then outline a temporal split followed by fitting imputation and feature-processing steps on training data only.
Official references
What to confirm
Distinguish leakage from future information from contamination caused by fitting preprocessing with validation or test data.
Key points
- Future information
- Time split
- Fit preprocessing
Notes
- Environment: Reference review of official documentation and hypothetical examples; no AWS resources or model training are required.
- A temporal split cannot prevent leakage if unavailable features remain. Check both feature-availability times and preprocessing after the split.
- With cross-validation, fit preprocessing within each training fold. Distinguish applying a fitted transformation from learning it again on evaluation data.
Foundation review
Data leakage
Information unavailable at prediction time that accidentally enters training or evaluation.
Temporal split
Separating earlier records for training from later records for validation or testing.