A company has collected customer comments on its products, rating them as safe or unsafe, using decision trees. The training dataset has the following features: id, date, full review, full review…

Question

Options

• ADrop the test samples with missing full review text fields, and then run through the test set.
• BCopy the summary text fields and use them to fill in the missing full review text fields, and then
• CUse an algorithm that handles missing data better than decision trees.
• DGenerate synthetic data to fill in the fields that are missing data, and then run through the test set.

• A
  4% (1)
• B
  83% (19)
• C
  4% (1)
• D
  9% (2)

Explanation

Option B is correct because the full review summary is a semantic proxy for the full review text - it was derived from the same content and captures the key meaning needed for classification. Using it to fill in the missing field preserves the test sample and provides the model with the most relevant substitute without discarding data or changing the pipeline.

Why the distractors fail:

• A is wrong because dropping test samples is unacceptable in practice - in production, you still need to classify every incoming review, and discarding records defeats the purpose of the model.
• C is wrong because switching algorithms is a disproportionate architectural change that doesn't solve the root problem; the question constrains you to the existing decision tree setup.
• D is wrong because generating synthetic full reviews is unnecessarily complex, costly, and risks introducing noise - especially when a real, semantically equivalent field (the summary) already exists.

Memory tip: Ask yourself "Is there a field already in the data that captures the missing field's meaning?" If yes, use it for imputation before considering dropping records, changing models, or generating synthetic data. When a summary exists, it's always the most grounded substitute for the full text.

No community discussion yet for this question.