The Role of Data in Effective AI Systems AIE Practice Questions (Page 4)

Questions 16–20

1. 16foundation · easy

   Which of the following is a common source of bias in AI training data?

   • AStoring data in a cloud database
   • BUsing a dataset that underrepresents certain demographic groups
   • CUsing a large number of training examples
   • DApplying normalization to all numeric features
   Reveal Solution Discussion Select an answer first
2. 17application · medium

   A retail company is building a churn-prediction model from customer records. The data contains duplicate customer IDs, inconsistent date formats, and missing values for income. The team wants to ensure the model learns reliable patterns. Which combination of actions best addresses the data quality issues before training?

   • AApply normalization to all numeric features and then train the model on the raw data.
   • BRemove duplicates, standardize date formats, and impute missing income values using a validated method.
   • CIncrease the dataset size by adding synthetic records to compensate for missing income values.
   • DDrop all records with missing income values and keep the original date formats.
   Reveal Solution Discussion Select an answer first
4. 18foundation · easy

   Which stage of the data lifecycle involves defining how long data will be kept and when it should be removed?

   • AArchival and deletion
   • BCollection
   • CStorage
   • DUsage
   Reveal Solution Discussion Select an answer first
5. 19application · medium

   A company is building a spam-detection model for email. They have a large dataset of emails, but only a small portion is labeled as spam or not spam. The labeling is done by a third-party service, and the company is concerned about label quality. What is the best way to ensure the labels are reliable?

   • ATrust the third-party service's labels because they are professionals and have experience in email classification.
   • BIncrease the number of emails labeled by the third-party service to dilute the impact of any individual error.
   • CUse the labels as-is and rely on the model to ignore any incorrect labels during training.
   • DHave multiple annotators label a subset of emails and measure agreement; resolve disagreements through discussion or a senior reviewer.
   Reveal Solution Discussion Select an answer first
6. 20application · medium

   An HR department is building a resume-screening model. The training data is from the past five years and contains far more resumes from male candidates than female candidates. The model is expected to shortlist candidates for interviews. What is the most appropriate action to reduce the risk of biased outcomes?

   • AUse a more complex model to capture subtle patterns in the imbalanced data.
   • BTrain the model on the current data and audit the outcomes after deployment.
   • CRemove the gender attribute from the dataset and train the model as-is.
   • DBalance the training data by collecting more resumes from underrepresented groups or using resampling techniques.
   Reveal Solution Discussion Select an answer first