Bias Detection CRAGE Practice Questions (Page 4)

Questions 16–20

1. 16application · medium

   A data science team is debugging a model that predicts customer churn. The model's predictions vary significantly when retrained on slightly different random samples of the same dataset. The team is concerned about bias. Which concept best describes this issue?

   • AVariance
   • BConfirmation bias
   • CSelection bias
   • DNoise
   Reveal Solution Discussion Select an answer first
2. 17foundation · easy

   Which fairness metric requires that the true positive rate and false positive rate are equal across groups?

   • ADemographic parity
   • BEqualized odds
   • CPredictive parity
   • DIndividual fairness
   Reveal Solution Discussion Select an answer first
4. 18application · medium

   A model predicts whether a job applicant will be a high performer. The team wants to ensure that among applicants who would actually be high performers, the model recommends them at the same rate across demographic groups. Which fairness metric should they use?

   • AEqual opportunity
   • BF1 score
   • CCalibration
   • DDemographic parity
   Reveal Solution Discussion Select an answer first
5. 19application · medium

   A model predicts credit risk. The fairness audit shows that the false positive rate is 0.3 for one ethnic group and 0.1 for another, while the false negative rate is equal across groups. The team must choose a mitigation strategy. Which interpretation of the results is most appropriate?

   • AThe model satisfies equalized odds because false negative rates are equal.
   • BThe model satisfies demographic parity because the overall approval rates are similar.
   • CThe model violates equalized odds because false positive rates differ, so mitigation should focus on reducing false positives for the disadvantaged group.
   • DThe model has no bias because false negative rates are equal, so no mitigation is needed.
   Reveal Solution Discussion Select an answer first
6. 20application · medium

   A model is being developed to predict insurance fraud. The data scientists notice that the model performs differently for customers from different regions. They suspect the difference is due to bias, but they are not sure whether it is data bias or algorithm bias. Which approach would best help them identify the source?

   • ARemove the region feature from the model entirely.
   • BTrain the model on a balanced dataset that has been reweighted to equalize regional representation, and compare performance across regions.
   • CIncrease the model's complexity to capture more regional nuances.
   • DUse a different algorithm that is known to be less biased.
   Reveal Solution Discussion Select an answer first