Manage AI/ML Transparency (e.g., Data Selection, Algorithm Selection) PMI-CPMAI Practice Questions (Page 6)

Questions 26–30

1. 26foundation · easy

   When recording the decision rationale for selecting an AI/ML model, what should be captured?

   • AThe reasoning behind the choice and the alternatives that were considered
   • BThe exact hyperparameters of the final model only
   • CThe names of the data scientists who approved the model
   • DThe cost of training the model in dollars
   Reveal Solution Discussion Select an answer first
2. 27foundation · easy

   Which of the following is an example of transparent reporting on data sources?

   • AListing the model's hyperparameters
   • BStating that the model was trained on a large dataset
   • CProviding the model's accuracy on a test set
   • DDocumenting that the training data was collected from customer surveys conducted in 2023
   Reveal Solution Discussion Select an answer first
4. 28application · medium

   A financial services firm replaced a legacy credit-scoring model with a new gradient boosting model. The new model improved the default rate prediction by 15% but is less transparent. During a regulatory review, the examiner asks why the firm abandoned the previous model. The firm's model governance documentation must capture this decision. What is the MOST important element to include in the decision record?

   • AThe names of the data scientists who trained the new model
   • BThe exact hyperparameters of the new gradient boosting model
   • CThe quantitative performance improvement and the accepted trade-off of reduced interpretability
   • DThe list of data sources used to train the legacy model
   Reveal Solution Discussion Select an answer first
5. 29application · medium

   A ride-sharing company uses a model to set dynamic pricing. A driver complains that their earnings dropped unexpectedly. The driver wants to know why. The company has a SHAP explanation for the driver's specific earnings prediction. What is the MOST appropriate way to communicate this to the driver?

   • ATell the driver the model is proprietary and cannot be explained
   • BSend the driver the raw SHAP values for all features in the model
   • CExplain that the drop was driven by a few key factors, such as increased driver supply in their area and a decrease in ride demand
   • DProvide the driver with the model's overall accuracy metric
   Reveal Solution Discussion Select an answer first
6. 30expert · hard

   A manufacturing company is selecting a predictive maintenance model. Model A (a deep neural network) has the highest accuracy (98%) but is a black box. Model B (a gradient boosting model) has 95% accuracy and provides feature importance. Model C (a logistic regression) has 91% accuracy but is fully interpretable. The maintenance team needs to justify downtime decisions to safety auditors, and the data science team has limited time to document the model. The company's policy requires that any model with accuracy below 93% must be rejected. Which model should the company select, and what is the PRIMARY justification?

   • AModel C, because it is fully interpretable, even though it fails the accuracy policy
   • BModel A, because the data science team can apply SHAP to make it interpretable after deployment
   • CModel A, because its 98% accuracy is the highest and the policy only mandates a minimum accuracy threshold
   • DModel B, because it meets the accuracy threshold and provides feature importance for the auditors
   Reveal Solution Discussion Select an answer first