What is the best method we have learned so far to handle missing data in a data set?

• A. Delete records
• B. Replace missing data with the mean of an attribute
• C. Replace missing data with the mode of an attribute
• D. Predict missing values with regression

Answer: (C)

Handling missing data by filling in with the most frequent value (the mode) is a simple, practical baseline approach. It uses an actual observed value, so you don’t introduce unfamiliar or unrealistic numbers. For categorical attributes, the mode preserves the existing category distribution and keeps records usable without discarding data. It also avoids distorting relationships as much as more aggressive imputations might—like predicting a numeric value that could create artificial patterns.

Deleting records reduces the dataset size and can bias results if the missingness isn’t random. Replacing with a mean or median works for numeric attributes but can distort the distribution of the feature and reduce variability. Regression imputation predicts values based on other attributes, which can be powerful but introduces modeling assumptions and the risk of leaking information or overfitting, especially in a simple, early-stage approach.