The Role of Data in Effective AI Systems AIE Practice Questions (Page 8)

Questions 36–40

1. 36application · medium

   A retail company is building a demand-forecasting model using three years of historical sales data. During preprocessing, the data science team discovers that the last six months of data have a different format for the 'store_id' field (e.g., 'NY-001' vs '001'), and that some stores have missing sales records for entire weeks. The team wants to ensure the model learns accurate patterns. What should the team do first?

   • ARemove all records with missing sales weeks and keep only the stores with complete data, ignoring the format issue.
   • BTrain the model on the data as-is, because modern algorithms can handle inconsistent formats and missing values automatically.
   • CIncrease the volume of data by adding more recent months of sales data to compensate for the missing weeks.
   • DStandardize the store_id format and decide how to handle missing weekly records, such as imputation or exclusion, before training.
   Reveal Solution Discussion Select an answer first
2. 37expert · hard

   A hospital is building an AI model to predict patient readmission risk using electronic health records (EHR). The data contains sensitive health information and is subject to HIPAA. The hospital wants to share the data with a research partner for model development. What is the most appropriate way to share the data while protecting patient privacy?

   • AEncrypt the data and share the encryption key with the partner so they can decrypt it for analysis.
   • BAggregate the data to the hospital level and share only summary statistics, losing all patient-level detail.
   • CShare the raw EHR data with the partner under a data use agreement, since the partner is also a healthcare entity.
   • DDe-identify the data by removing direct identifiers and applying a formal privacy model like differential privacy before sharing.
   Reveal Solution Discussion Select an answer first
4. 38application · medium

   An autonomous vehicle company is labeling road images for training a pedestrian-detection model. The labeling team is working in shifts, and the company wants to ensure high-quality labels. Which practice is most important to maintain labeling consistency?

   • AAutomatically label all images using a pre-trained model and skip manual labeling entirely.
   • BDevelop a detailed labeling guideline with examples and edge cases, and conduct regular inter-annotator agreement checks.
   • CAllow each annotator to use their own judgment to label images quickly, as long as they meet a daily quota.
   • DUse the same annotator for all images to ensure consistency, even if it slows down the process.
   Reveal Solution Discussion Select an answer first
5. 39application · medium

   A bank is developing a credit-scoring model. The dataset contains customer information collected from multiple branches, and the data team notices that some fields have conflicting values (e.g., the same customer has different income values in different branches). What is the most important data quality dimension to address?

   • AAccuracy, because the income values might be incorrect.
   • BTimeliness, because the data must be current to reflect the customer's latest income.
   • CCompleteness, because missing income values need to be imputed.
   • DConsistency, because conflicting values across sources must be reconciled before training.
   Reveal Solution Discussion Select an answer first
6. 40expert · hard

   A data team is preparing a dataset for a deep learning model. The dataset contains a mix of categorical and numerical features, and some features have very different scales. The team also notices that some categorical features have rare categories that appear only a few times. What is the most appropriate preprocessing strategy?

   • AEncode categorical features with a method that handles rare categories (e.g., grouping rare categories) and normalize numerical features.
   • BApply one-hot encoding to all categorical features without grouping rare categories.
   • CRemove all categorical features and use only numerical features.
   • DNormalize categorical features by converting them to numerical codes.
   Reveal Solution Discussion Select an answer first