A fraud-detection training set has many missing fraud labels for one region.
The production model should have similar performance across regions.
What should be done first?
Investigate the distribution and cause of missing labels, then choose imputation, exclusion, or additional collection based on measured impact.
Detailed explanation
A regional pattern may create bias, so the cause and model impact should be assessed before processing.
This may remove important regional representation and create another bias.
A uniform value can introduce systematic label bias.
Temperature affects generative inference randomness and does not repair missing training labels.
Check official documentation
Compare the following points with the official documentation.
AWS公式AIF-C01ガイドのDomain 1.3でデータ品質、代表性、前処理の説明を確認する。
Official references
What to confirm
欠損処理を原因・分布・性能影響の評価と結び付けて説明できる。
Key points
- Missingness bias
- Data representativeness
- Processing policy
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
Missingness bias
When missing values occur more often for some groups or conditions, the learned result can become uneven.
Imputation
Replacing a missing value with an estimate or explicit indicator while preserving information about its absence.