Production accuracy has decreased.
Investigate input distribution changes and label quality separately.
Choose two investigation steps.
Compare production and training inputs, missingness, units, and category ratios, and separately check label delays, errors, definition changes, and model versions.
Detailed explanation
A root-cause investigation is missing.
These checks can reveal data drift.
The cause may be data or labels rather than the model.
Quality causes can be separated from input drift.
Check official documentation
Compare the following points with the official documentation.
AWS公式AIF-C01 Domain 1.2のデータドリフト、ラベル品質、モデル監視を確認する。
Official references
What to confirm
モデル劣化をデータ・ラベル・モデルの仮説に分解して調査できる。
Key points
- Drift
- Label quality
- Root cause
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
Data drift
A change in the distribution or statistics of production inputs over time.
Label drift
A change in the prevalence, definition, or labeling process for target outcomes.