Objective 4.1 Prepare Data Sets for Modeling CERTIFIED-DATA-SCIENCE-PRACTITIONER Practice Questions (Page 4)

Questions 16–20

1. 16application · easy

   A data science team is collaborating on a project and each member runs the same data preparation script. They notice that they get different training and test sets each time the script is run, even on the same data. What is the most likely cause?

   • AThe script does not set a fixed random seed before splitting.
   • BThe script shuffles the data before splitting.
   • CThe script uses temporal splitting.
   • DThe script uses stratified splitting.
   Reveal Solution Discussion Select an answer first
2. 17application · medium

   A data scientist is working with a dataset where each row represents a single user session, but multiple sessions belong to the same user. They want to split the data to prevent the same user from appearing in both training and test sets. Which splitting approach is most appropriate?

   • AGroup-based split by user ID
   • BRandom split by session ID
   • CStratified split by session length
   • DTemporal split by session timestamp
   Reveal Solution Discussion Select an answer first
4. 18application · medium

   A team is building a model to predict stock prices using historical daily data. They split the data randomly into training and test sets. What is the most likely problem with this approach?

   • AThe model will be trained on future data and evaluated on past data, causing leakage.
   • BThe class distribution will be imbalanced in the test set.
   • CThe test set will be too small to evaluate the model.
   • DThe model will not be able to learn from the data.
   Reveal Solution Discussion Select an answer first
5. 19application · medium

   A data scientist has a small dataset of 500 records for a binary classification problem. They need to train a model and get a reliable estimate of performance. Which splitting strategy is most appropriate?

   • AUse a 90/10 train-test split
   • BUse k-fold cross-validation (e.g., 5-fold) on the entire dataset
   • CUse a 70/30 train-test split with a fixed seed
   • DUse a 50/50 train-test split
   Reveal Solution Discussion Select an answer first
6. 20foundation · easy

   What is a key advantage of random splitting over non-random splitting for a typical cross-sectional dataset?

   • AIt guarantees the class distribution is preserved
   • BIt always produces the same split every time
   • CIt reduces the risk of bias from any inherent order in the data
   • DIt is the only method that works for large datasets
   Reveal Solution Discussion Select an answer first