Objective 4.3 Maintain the Model Postproduction AIP-210 Practice Questions (Page 4)

Questions 16–20

1. 16expert · hard

   An insurance company has a model that sets premiums. The model was developed using data that included a protected attribute, and an audit revealed that the model is biased against a certain demographic. The company must fix the bias while maintaining the model's overall accuracy. What is the BEST approach?

   • ARetrain the model with bias mitigation techniques and document the changes for compliance.
   • BRemove the protected attribute from the model's input features and retrain.
   • CAdjust the model's decision threshold to reduce false positives for the affected group.
   • DRetire the model and develop a new one from scratch without using any sensitive data.
   Reveal Solution Discussion Select an answer first
2. 17application · medium

   A logistics company deploys a delivery time prediction model. The model's performance is monitored using a dashboard that tracks mean absolute error (MAE). The MAE has been stable for months, but the company's customer satisfaction scores have declined. The team suspects the model is making systematically biased predictions for a specific region. What monitoring approach would BEST capture this issue?

   • ACompare the model's predictions to the previous year's delivery times.
   • BMonitor the model's version number to ensure the latest version is deployed.
   • CSegment the MAE by region and compare it to the overall MAE.
   • DIncrease the frequency of retraining to weekly.
   Reveal Solution Discussion Select an answer first
4. 18application · medium

   A ride-sharing company's surge pricing model is retrained monthly. After a major holiday, the model's performance degrades because the demand pattern was unusual. The team expects demand to return to normal within a few weeks. What is the BEST course of action?

   • AIncrease the retraining frequency to daily for the next month.
   • BRoll back to the previous model version until demand normalizes.
   • CWait for the next scheduled retraining and continue monitoring to see if performance recovers.
   • DRetrain the model immediately using only the holiday data to capture the new pattern.
   Reveal Solution Discussion Select an answer first
5. 19foundation · easy

   What is the main purpose of retraining a model with newly collected data?

   • ATo maintain or improve prediction accuracy as real-world patterns evolve
   • BTo reduce the model's inference latency
   • CTo reduce the size of the training dataset
   • DTo eliminate the need for ongoing model monitoring
   Reveal Solution Discussion Select an answer first
6. 20foundation · easy

   What is the primary purpose of comprehensive model documentation?

   • ATo ensure the model is deployed to production automatically
   • BTo replace the need for model monitoring
   • CTo enable others to understand, reproduce, and maintain the model
   • DTo increase the model's prediction speed
   Reveal Solution Discussion Select an answer first