Split an imbalanced classification dataset.
Keep class proportions from becoming extreme in each split.
Choose two appropriate practices.
Consider stratification while also checking time order, group overlap, leakage, and recording the split conditions.
Detailed explanation
Results cannot be reproduced or explained.
Minority examples are more likely to appear in each fold.
Time-aware splits may be required to prevent future leakage.
Stratification alone does not solve leakage or grouping.
Check official documentation
Compare the following points with the official documentation.
AWS公式AIF-C01 Domain 1.3の評価データ分割と不均衡データを確認する。
Official references
What to confirm
層化分割の利点と、適用できない条件を説明できる。
Key points
- Stratification
- Class balance
- Time
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
Stratified split
Partitioning data while preserving the proportion of selected classes in each subset.
Group split
Keeping records from the same customer, device, or entity in one subset.