Objective 2.5 Address Business Risks, Ethical Concerns, and Related Concepts in Data Exploration/feature Engineering AIP-210 Practice Questions (Page 4)

Questions 16–20

1. 16application · medium

   A data science team discovers that a feature used in a loan-approval model is derived from a data source that has known inaccuracies for a particular demographic group. The team needs to communicate this risk to the business stakeholders who are not technical. What is the most effective way to convey the issue?

   • APresent a chart showing the model's approval rates by demographic group, highlighting the potential for unfair outcomes.
   • BState that the model is 'fine' because the overall accuracy is high.
   • CExplain the technical details of the data source's schema and the imputation method used.
   • DSend a detailed email with the full data lineage report and ask stakeholders to review it.
   Reveal Solution Discussion Select an answer first
2. 17foundation · easy

   While exploring a dataset, a data scientist notices that a column contains personally identifiable information (PII) such as email addresses. What business risk is most directly associated with this observation?

   • ARisk of privacy violations if the data is not properly protected
   • BRisk of data leakage from the target variable
   • CRisk of biased predictions due to skewed class distribution
   • DRisk of model overfitting due to high cardinality
   Reveal Solution Discussion Select an answer first
4. 18application · medium

   A healthcare startup is building a patient-readmission prediction model. During data exploration, a data scientist discovers that the 'insurance provider' field contains free-text entries with inconsistent formatting (e.g., 'Aetna', 'aetna', 'AETNA Inc.'). The team plans to engineer a categorical feature from this field. What is the most appropriate first step to address the business risk associated with this finding?

   • ADrop the 'insurance provider' field entirely to avoid introducing any bias into the model.
   • BEncrypt the 'insurance provider' field before feature engineering to protect patient privacy.
   • CCreate a binary feature indicating whether the text is in uppercase to capture formatting differences.
   • DStandardize the text values using case normalization and string matching before creating the categorical feature.
   Reveal Solution Discussion Select an answer first
5. 19expert · hard

   A hiring platform develops a model to rank job applicants. The team engineers a feature called 'years of experience' and finds that the model has a disparate impact on older workers. They try removing the feature, but the model still shows bias because 'number of previous jobs' correlates with age. Which approach is most likely to mitigate the bias while maintaining model performance?

   • AApply a post-processing technique that adjusts the model's decision threshold for different age groups to equalize outcomes.
   • BUse an adversarial debiasing technique during model training that penalizes the model for using information about age.
   • CRemove all features that correlate with age, including 'number of previous jobs'.
   • DRe-engineer the 'years of experience' feature to be a categorical band (e.g., 0-5, 6-10, 10+ years) to reduce granularity.
   Reveal Solution Discussion Select an answer first
6. 20application · medium

   A data scientist at a marketing firm is exploring a dataset containing customer purchase history to engineer features for a recommendation model. The dataset includes full names, email addresses, and purchase amounts. The firm wants to allow the data scientist to work with the data while minimizing privacy risks. What is the most appropriate technique to apply before feature engineering?

   • ATokenize the email addresses and names using a one-way hash function.
   • BRemove the purchase amount column to reduce the sensitivity of the data.
   • CEncrypt the entire dataset with a symmetric key and share the key with the data scientist.
   • DAggregate the data to the customer segment level before exploration.
   Reveal Solution Discussion Select an answer first