Handle Missing and Categorical Data SNOWPRO-ADVANCED-DATA-SCIENTIST Practice Questions (Page 2)

Questions 6–10

1. 6expert · hard

   A data scientist is building a model to predict customer satisfaction. The dataset has a 'store_id' feature with 2,000 unique values. Some stores have very few transactions (e.g., 5 records), while others have thousands. The team wants to use target encoding. What is the most appropriate way to handle the stores with very few records?

   • AApply smoothing by blending the store's target mean with the global mean, weighted by the number of records.
   • BUse the raw target mean for each store, regardless of sample size.
   • CGroup stores with few records into a single 'Other' category before encoding.
   • DDrop all stores with fewer than 50 records.
   Reveal Solution Discussion Select an answer first
2. 7foundation · easy

   In a clinical trial dataset, some patients have missing blood pressure readings because the measurement device malfunctioned at random times, unrelated to patient characteristics or the blood pressure value itself. Which type of missingness is this?

   • AMissing completely at random (MCAR)
   • BStructurally missing
   • CMissing at random (MAR)
   • DMissing not at random (MNAR)
   Reveal Solution Discussion Select an answer first
4. 8application · medium

   A data scientist is building a linear regression model to predict employee salary. The dataset includes a 'department' feature with 20 unique values, and a 'job_level' feature with 5 ordered levels (e.g., Junior, Mid, Senior, Lead, Manager). What is the most appropriate encoding strategy?

   • AOne-hot encode both 'department' and 'job_level'.
   • BLabel encode both 'department' and 'job_level'.
   • COne-hot encode 'department' and ordinal encode 'job_level'.
   • DOrdinal encode 'department' and one-hot encode 'job_level'.
   Reveal Solution Discussion Select an answer first
5. 9application · medium

   A data scientist is building a neural network model to predict customer satisfaction. The dataset includes a 'country' feature with 50 unique values. The team wants to use an embedding layer. What is the most appropriate way to prepare the 'country' feature?

   • AOne-hot encode 'country' and feed it directly into the network.
   • BDrop the 'country' feature.
   • CUse frequency encoding and feed it as a single numeric feature.
   • DLabel encode 'country' and feed it into an embedding layer.
   Reveal Solution Discussion Select an answer first
6. 10expert · medium

   A data scientist is building a tree-based model (Random Forest) to predict loan approval. The dataset includes a 'credit_score_band' feature with ordered categories: Poor, Fair, Good, Excellent. The team is debating between ordinal encoding and one-hot encoding. What is the most appropriate choice and why?

   • AFrequency encoding, because it captures the distribution of categories.
   • BOrdinal encoding, because it preserves the natural order of the categories.
   • COne-hot encoding, because it avoids imposing a false order.
   • DBinary encoding, because it reduces dimensionality while preserving order.
   Reveal Solution Discussion Select an answer first