Manage AI/ML Transparency (e.g., Data Selection, Algorithm Selection) PMI-CPMAI Practice Questions (Page 4)

Questions 16–20

1. 16expert · hard

   A telecom company is deploying a churn prediction model. The stakeholders are: (1) marketing managers who want to know which customer segments to target, (2) data protection regulators who need to verify the model does not discriminate, and (3) customer service agents who need to explain retention offers to individual customers. The company has limited budget and can only implement ONE interpretability tool. Which tool BEST serves the widest range of stakeholder needs?

   • ASHAP, because it provides both local explanations for individual predictions and global feature importance summaries
   • BFeature importance from the model itself, because it is the cheapest to compute
   • CA decision tree surrogate model, because it is simple to explain to all stakeholders
   • DLIME, because it provides local explanations for individual customer decisions
   Reveal Solution Discussion Select an answer first
2. 17foundation · easy

   Why is it important to document data preprocessing steps in a transparent report?

   • ATo reduce the time needed for model training
   • BTo increase the model's accuracy
   • CSo that stakeholders can understand how the data was transformed and assess the impact on model outcomes
   • DTo comply with data storage regulations
   Reveal Solution Discussion Select an answer first
4. 18application · medium

   A healthcare startup is building a model to predict patient readmission risk. The data science team evaluated logistic regression, random forest, and gradient boosting. All three achieved similar AUC scores, but the team selected logistic regression because the hospital's compliance officers require a clear explanation of how each clinical feature contributes to a prediction. The startup must document this decision for an upcoming audit. What should the documentation emphasize as the PRIMARY selection criterion?

   • AThe random forest model's ability to handle missing data without imputation
   • BThe logistic regression model's superior predictive accuracy on the validation set
   • CThe gradient boosting model's faster training time on the available hardware
   • DThe regulatory requirement for interpretability, which logistic regression satisfied better than the ensemble methods
   Reveal Solution Discussion Select an answer first
5. 19application · medium

   A retail company is building a customer churn model. The training data combines purchase history from the company's CRM, demographic data purchased from a third-party vendor, and social media sentiment scores. The company must publish a transparency report for its data protection regulator. What is the MINIMUM information the report must include to be considered transparent about data sources?

   • AThe total number of records in the training dataset
   • BThe SQL queries used to extract the data from the CRM
   • CThe name of the third-party vendor and the date the data was purchased
   • DA description of each data source, its original purpose, and how consent or legal basis applies to its use
   Reveal Solution Discussion Select an answer first
6. 20application · medium

   A media company is building a content recommendation model. The team evaluated a collaborative filtering model and a content-based filtering model. The collaborative filtering model performed better in offline tests but is harder to explain. The content-based model is simpler and easier to explain to content partners. The team selects the content-based model. What should the decision record include to be complete?

   • AOnly the final selection and the performance metrics of the content-based model
   • BThe names of the team members who voted for each model
   • CThe performance metrics of both models and the rationale that explainability to content partners outweighed the performance difference
   • DThe code used to train the content-based model
   Reveal Solution Discussion Select an answer first