Which of the following strategies used by modelers to enhance data accuracy has the GREATEST risk of bias and information loss?

Question

Options

• AFilling blank attributes in records with the mean, median, or mode within a grouping
• BIdentifying and deleting duplicate entries in the data set
• CSeparating multiple data attributes within one field into individual attribute columns
• DPlacing numerical data into bins or buckets for a manageable quantity of correlations and result

• A
  44% (14)
• B
  31% (10)
• C
  9% (3)
• D
  16% (5)

Explanation

Imputation-filling missing attribute values with the mean, median, or mode (A)-carries the greatest risk of bias and information loss. This technique assumes missing data follows the same distribution as observed data, which may not be true (especially when data is missing not at random). It artificially inflates the dataset with synthetic values and can skew model outputs toward the dominant patterns of the group used for imputation. Removing duplicates (B), splitting multi-value fields (C), and binning numerical data (D) are generally safer preprocessing steps with less inherent risk of introducing systematic bias.

No community discussion yet for this question.