Objective 4.3 Maintain the Model Postproduction AIP-210 Practice Questions (Page 6)

Questions 26–30

1. 26foundation · easy

   What is the primary trigger for retraining a machine learning model in production?

   • AA change in the model's training code
   • BAn increase in the number of users accessing the model
   • CA scheduled calendar date, regardless of model performance
   • DA detected decline in model performance or significant data drift
   Reveal Solution Discussion Select an answer first
2. 27foundation · easy

   Which metric is most commonly used to detect data drift in a production model?

   • AModel accuracy on the training set
   • BThe time taken to serve each prediction
   • CThe statistical distance between training and production data distributions
   • DThe number of model retraining cycles
   Reveal Solution Discussion Select an answer first
4. 28expert · hard

   A manufacturing company uses a predictive maintenance model. The model's performance is monitored, and the team has set an alert if accuracy drops below 90%. The alert has not fired, but the maintenance team reports that the model is missing many failures. The team investigates and finds that the model's recall has dropped significantly, but accuracy remains high because failures are rare. What should the team do?

   • ARoll back to the previous model version that had higher recall.
   • BIncrease the alert threshold for accuracy to 95% to catch the issue earlier.
   • CAdd a recall-specific alert to the monitoring system and retrain the model if recall drops below a threshold.
   • DRetrain the model with more data to improve overall accuracy.
   Reveal Solution Discussion Select an answer first
5. 29foundation · easy

   What is the primary purpose of model governance?

   • ATo manage the model lifecycle with policies that ensure compliance and auditability
   • BTo ensure models are deployed as quickly as possible
   • CTo maximize the accuracy of every model
   • DTo reduce the cost of model training
   Reveal Solution Discussion Select an answer first
6. 30expert · hard

   A hospital uses a model to predict patient readmission risk. The model was trained on data from a specific patient population. After a merger, the hospital now serves a more diverse population. The model's performance has declined, but the decline is not uniform across all patient groups. The hospital's ethics board requires that the model not discriminate against any group. What should the team do?

   • ASet a single performance threshold for all groups to simplify monitoring.
   • BRetrain the model on the combined patient population to improve overall accuracy.
   • CRoll back to the previous model version that was trained on the original population.
   • DMonitor performance metrics separately for each patient group and retrain the model if any group's performance degrades.
   Reveal Solution Discussion Select an answer first