Oversee Data Evaluation PMI-CPMAI Practice Questions (Page 5)

Questions 21–25

1. 21application · medium

   A data engineer is reviewing a dataset for a churn prediction model. The dataset has a 'customer_id' column with unique values, a 'churn' column with values 'Yes' and 'No', and a 'monthly_charges' column with numeric values. The modeling team plans to use a logistic regression algorithm. Which schema issue would most likely require attention before modeling?

   • AThe 'customer_id' column has too many unique values and should be dropped.
   • BThe 'monthly_charges' column needs to be normalized to a 0-1 range.
   • CThe dataset lacks a timestamp column for time-based splitting.
   • DThe 'churn' column is categorical and must be encoded as a binary numeric variable.
   Reveal Solution Discussion Select an answer first
2. 22foundation · easy

   What is the main goal of exploratory data analysis (EDA) in the context of AI model development?

   • ATo deploy the model to production
   • BTo secure the data pipeline
   • CTo understand data characteristics and patterns
   • DTo write the final model code
   Reveal Solution Discussion Select an answer first
4. 23expert · hard

   A retail company is building a real-time inventory management model. The model uses point-of-sale (POS) data, which is streamed continuously, and supplier data, which is updated weekly. The team is evaluating the data for training and inference. What is the most critical data evaluation concern?

   • AThe supplier data should be discarded because it is not real-time.
   • BThe supplier data may be stale for real-time decisions, but it is still relevant for weekly restocking predictions.
   • CThe POS data is too fresh and may cause overfitting.
   • DThe schema of the POS data is incompatible with the supplier data.
   Reveal Solution Discussion Select an answer first
5. 24expert · hard

   During EDA of a customer transaction dataset, a data scientist discovers that the 'transaction_amount' column has a large number of zero values (30%). The business context is that zero-amount transactions are legitimate (e.g., refunds or adjustments). The model will predict customer lifetime value. What should the data scientist do?

   • AIgnore the zeros because they are a small fraction of the data.
   • BRemove all zero-amount transactions to avoid skewing the distribution.
   • CKeep the zero values as they are legitimate, but document the distribution and consider a two-stage model (one for probability of purchase, one for amount).
   • DReplace zero values with a small positive number to avoid log transformation issues.
   Reveal Solution Discussion Select an answer first
6. 25application · medium

   An AI team is building a model to predict loan default. The training dataset contains 95% non-default and 5% default records. The team plans to train the model on this raw distribution. What is the primary data evaluation concern with this approach?

   • AThe model may become biased toward the majority class and perform poorly on default cases.
   • BThe data is not fresh enough to reflect current economic conditions.
   • CThe schema is incompatible with the modeling algorithm.
   • DThe data may contain duplicate records that inflate the sample size.
   Reveal Solution Discussion Select an answer first