Bias Detection CRAGE Practice Questions (Page 9)

Questions 41–42

1. 41expert · hard

   A model is used to recommend treatment plans for patients. The team must ensure that the model does not discriminate against any ethnic group. They compute several fairness metrics and find that the model satisfies demographic parity and equalized odds, but calibration is poor for one ethnic group. What is the most appropriate action?

   • ARemove the ethnic group feature from the model.
   • BInvestigate the calibration disparity because poor calibration can lead to incorrect treatment decisions for that group.
   • CSwitch to a different fairness metric that is easier to satisfy.
   • DIgnore the calibration issue because the other metrics are satisfied.
   Reveal Solution Discussion Select an answer first
2. 42foundation · easy

   Which of the following is a quantitative method for detecting bias in a dataset?

   • AReviewing the model's documentation
   • BConducting stakeholder interviews
   • CCalculating the distribution of a protected attribute across the dataset
   • DObserving user interactions with the model
   Reveal Solution Discussion Select an answer first