Which technique is most effective for handling missing values in a small dataset with non-random missingness?

Question

Data Science — Medium

Which technique is most effective for handling missing values in a small dataset with non-random missingness?

Accepted Answer

When data is not missing at random (MNAR) or missing at random (MAR), simple imputation methods like mean or constant replacement introduce significant bias. Multiple imputation (such as IterativeImputer) models each feature with missing values as a function of other features, preserving the statistical relationships and uncertainty better than single-point imputation methods.