Handle Missing and Categorical Data SNOWPRO-ADVANCED-DATA-SCIENTIST Practice Questions (Page 1)

Questions 1–5

1. 1expert · hard

   A data scientist is analyzing a survey dataset where respondents could skip questions. The 'income' question was skipped by 25% of respondents. Analysis shows that younger respondents (age < 30) are significantly more likely to skip the income question. The 'age' variable is fully observed. The team plans to build a model to predict spending behavior. What is the most appropriate characterization of the missingness and the best handling strategy?

   • AMNAR; use a selection model.
   • BMAR; use multiple imputation that includes 'age' as a predictor.
   • CMAR; impute with the mean of 'income'.
   • DMCAR; use listwise deletion.
   Reveal Solution Discussion Select an answer first
2. 2foundation · easy

   A data scientist is analyzing a survey dataset where respondents who earn above a certain income threshold are systematically less likely to report their income. Which type of missingness does this scenario describe?

   • AMissing at random (MAR)
   • BIgnorable missingness
   • CMissing completely at random (MCAR)
   • DMissing not at random (MNAR)
   Reveal Solution Discussion Select an answer first
4. 3expert · hard

   A data scientist is building a model to predict insurance claim severity. The dataset has a 'policy_type' feature with 500 unique values. The team is considering target encoding. What is the most important additional step to avoid overfitting?

   • AApply smoothing to the target encoding, such as using a weighted average with the global mean.
   • BIncrease the number of trees in the gradient boosting model.
   • CUse cross-validation to evaluate the model.
   • DUse frequency encoding instead.
   Reveal Solution Discussion Select an answer first
5. 4expert · hard

   A healthcare analytics team is building a model to predict patient readmission. The 'blood_pressure' field is missing for 30% of records. After investigation, they find that the missingness is not related to any observed patient characteristic, but they suspect that patients with extremely high blood pressure were less likely to have it recorded due to emergency triage protocols. The model will be deployed in the same hospital system. What is the most defensible approach?

   • AUse mean imputation, since the missingness is unrelated to observed variables.
   • BApply multiple imputation using all available features, assuming MAR.
   • CCreate a 'blood_pressure_missing' indicator and include it in the model without imputation.
   • DUse a model-based imputation that incorporates a 'missing not at random' assumption, such as a selection model, and validate with sensitivity analysis.
   Reveal Solution Discussion Select an answer first
6. 5foundation · easy

   Which encoding method replaces each category with the mean of the target variable for that category, and is often used for high-cardinality categorical features in supervised learning?

   • AOne-hot encoding
   • BLabel encoding
   • CFrequency encoding
   • DTarget encoding
   Reveal Solution Discussion Select an answer first