Oversee Model Governance PMI-CPMAI Practice Questions (Page 4)

Questions 16–20

1. 16application · medium

   An e-commerce recommendation model's click-through rate (CTR) has dropped 15% in the last two weeks. The input feature distributions look unchanged, but the model's predictions are less aligned with actual user behavior. The team suspects the relationship between features and outcomes has shifted. What should they do first?

   • ARetrain the model immediately with the most recent data to capture the new behavior.
   • BRun a statistical test comparing recent model predictions against actual outcomes to confirm model drift.
   • CIncrease the alert threshold so the drop is not flagged as an anomaly.
   • DRoll back to the previous model version to restore the original CTR.
   Reveal Solution Discussion Select an answer first
2. 17foundation · easy

   Which factor is most likely to trigger the need for an unscheduled retraining of a model?

   • AA significant drop in model accuracy detected by monitoring
   • BThe model's inference latency is stable
   • CThe model has been in production for a short time
   • DThe model's training data has not changed
   Reveal Solution Discussion Select an answer first
4. 18foundation · easy

   What is the primary purpose of implementing a model versioning system?

   • ATo ensure the model is deployed to the correct environment
   • BTo improve the model's accuracy by tuning hyperparameters
   • CTo automatically retrain the model when new data arrives
   • DTo uniquely identify and track different iterations of a model, including metadata and artifacts
   Reveal Solution Discussion Select an answer first
5. 19expert · hard

   A media company's content-recommendation model is updated frequently. The team maintains a model registry, but different teams use different naming conventions, and some models are overwritten without a clear audit trail. The governance team wants to ensure that every model version is uniquely identifiable and that its performance can be compared over time. What is the most important practice to implement?

   • AA centralized model registry that assigns a unique version ID to every model, with immutable metadata and artifacts.
   • BA shared spreadsheet that tracks model versions and their performance metrics.
   • CA policy that requires all teams to use the same naming convention for their models.
   • DA process where each team maintains its own model registry and shares results in a monthly meeting.
   Reveal Solution Discussion Select an answer first
6. 20expert · hard

   A hospital deploys a sepsis-prediction model that alerts clinicians. The model's AUC is high, but the clinical team reports that the alerts are often ignored because they are not actionable. The model's precision is low, generating many false alarms. The hospital wants to improve the model's usefulness without sacrificing recall for true sepsis cases. What is the best approach?

   • ALower the alert threshold to catch more true positives, increasing recall.
   • BAdjust the alert threshold to balance precision and recall based on the clinical cost of false positives versus false negatives, and monitor the impact.
   • CRetrain the model with a different algorithm to improve both precision and recall.
   • DRaise the alert threshold to reduce false alarms, accepting a lower recall.
   Reveal Solution Discussion Select an answer first