Feature Imputation

Question 1

Feature Imputation Techniques

Answer 1

• Nearest neighbor imputation (kNN)
• Drop the data
• Use mean or median

Question 2

Data Rows with Null Values Tradeoffs

Answer 2

• If you are confident the data is missing at random and you will have a large dataset remaining after the missing data is dropped, dropping data can be an effective solution
• If the data is not missing at random, then dropping the dataset can introduce significant bias into your model
• Dropping too much data is dangerous

Question 3

Dropping Columns with High Nullability

Answer 3

• Before dropping features, consider runing a feature importance analysis on the data you have

Question 4

Mean, Median, or Summary Statistic Substitution

Answer 4

• Generally, not a good approach. The more data is missing, the more harmful this approach is.
• Can introduce inconsistent bias