Manage AI/ML Transparency (e.g., Data Selection, Algorithm Selection) PMI-CPMAI Practice Questions (Page 5)

Questions 21–25

1. 21expert · hard

   A pharmaceutical company uses a machine learning model to predict drug side effects. The model is a complex ensemble. The company must provide explanations to two audiences: (1) the FDA, which requires a rigorous, mathematically sound explanation of the model's behavior, and (2) clinical trial nurses, who need simple, actionable reasons for why a specific patient might experience a side effect. The company has a limited budget and can implement only ONE interpretability method. Which method BEST satisfies both audiences?

   • ALIME, because it provides local explanations that nurses can use
   • BSHAP, because it provides both local explanations for individual patients and global feature importance with a solid theoretical foundation
   • CA single decision tree, because it is easy to explain to both audiences
   • DFeature importance from the model, because it is simple and cheap
   Reveal Solution Discussion Select an answer first
2. 22expert · hard

   A research institute is building a model to predict disease outbreaks using public health data. The raw data includes reports from hospitals, which contain a mix of structured fields (e.g., patient count) and unstructured notes. The data team applies a text-mining algorithm to extract symptoms from the notes, then merges this with the structured data. The team also applies a data augmentation technique to balance the dataset. For a peer-reviewed transparency report, what is the MOST important preprocessing detail to disclose?

   • AA complete pipeline description covering text extraction, the merge process, and the augmentation technique
   • BThe specific text-mining algorithm used to extract symptoms from unstructured notes
   • CThe data augmentation technique and the text-mining algorithm, but not the merge process
   • DThe total number of records in the final dataset
   Reveal Solution Discussion Select an answer first
4. 23foundation · easy

   Which of the following is an example of a stakeholder-specific explainability requirement?

   • ARequiring the same level of detail for all stakeholders
   • BAvoiding any explanation to prevent confusion
   • CFocusing only on technical performance metrics
   • DProviding end-users with a simple explanation of why a loan was denied
   Reveal Solution Discussion Select an answer first
5. 24foundation · easy

   What is the purpose of implementing model interpretability tools?

   • ATo make model predictions understandable to humans
   • BTo reduce the model's training time
   • CTo increase the model's accuracy
   • DTo automatically retrain the model
   Reveal Solution Discussion Select an answer first
6. 25application · medium

   A hospital uses a black-box model to predict patient deterioration. A physician questions a specific alert for a patient and wants to know why the model flagged this patient. The data science team applies LIME to this specific prediction. What does the LIME output provide to the physician?

   • AA simplified, interpretable explanation of which features (e.g., heart rate, blood pressure) contributed most to THIS patient's alert
   • BThe exact mathematical formula of the original black-box model
   • CA global ranking of the most important features across all patients in the hospital
   • DThe patient's raw vital signs from the electronic health record
   Reveal Solution Discussion Select an answer first