Objective 2.5 Address Business Risks, Ethical Concerns, and Related Concepts in Data Exploration/feature Engineering AIP-210 Practice Questions (Page 3)

Questions 11–15

1. 11foundation · easy

   When communicating risks to non-technical stakeholders, what is the most effective approach?

   • AAvoid discussing risks to prevent alarm
   • BProvide detailed statistical formulas
   • CUse technical jargon to ensure precision
   • DFocus on the business impact and use plain language
   Reveal Solution Discussion Select an answer first
2. 12application · medium

   A government agency is developing a model to allocate resources for public services. During feature engineering, the team discovers that the 'income' feature has a significant number of outliers that appear to be data entry errors. The team is concerned about the impact on the model. What is the most responsible approach to handle these outliers?

   • ACap all outliers at the 99th percentile to reduce their influence on the model.
   • BReplace outlier values with the median income to minimize their impact.
   • CInvestigate the source of the outliers to determine if they are genuine or errors, and document the decision to treat or remove them.
   • DRemove all rows with outlier income values to ensure the model is trained on clean data.
   Reveal Solution Discussion Select an answer first
4. 13expert · hard

   A bank is developing a credit-scoring model. The feature engineering team creates a 'postal code' feature by one-hot encoding all 40,000 postal codes in the country. During review, a compliance officer notes that postal code is strongly correlated with race and income. The team must balance model performance with regulatory fairness. Which approach best addresses the ethical concern while preserving model utility?

   • AReplace the high-cardinality postal code with a smaller set of region-level features engineered to capture relevant economic indicators while monitoring disparate impact.
   • BKeep the one-hot encoded postal codes because they are not explicitly protected attributes under the law.
   • CRemove all geographic features from the model to guarantee fairness.
   • DAdd a feature that explicitly encodes the racial composition of each postal code to allow the model to adjust for it.
   Reveal Solution Discussion Select an answer first
5. 14expert · hard

   A multinational corporation is building a global customer analytics model. The data science team needs to engineer features from customer data that includes personally identifiable information (PII) from multiple countries with different privacy regulations. The team wants to centralize the data for analysis. What is the most appropriate approach to balance utility and privacy compliance?

   • AKeep all data in its country of origin and only share aggregated statistics.
   • BCentralize all raw data in a single data lake and rely on access controls to protect it.
   • CApply a consistent pseudonymization technique to all PII before centralizing the data, and document the mapping separately.
   • DAnonymize the data by removing all direct identifiers and do not document the process.
   Reveal Solution Discussion Select an answer first
6. 15foundation · easy

   A data scientist creates a complex feature that improves model accuracy but is difficult for stakeholders to understand. Which ethical concern does this raise?

   • AData quality, because the feature may be noisy
   • BSecurity, because the feature may be vulnerable to attacks
   • CPrivacy, because the feature may contain personal data
   • DAccountability, because it becomes hard to explain why the model makes certain decisions
   Reveal Solution Discussion Select an answer first