What is the primary risk of using K-Nearest Neighbors (KNN) imputation when the feature space is not properly scaled?

Question

Data Science — Hard

What is the primary risk of using K-Nearest Neighbors (KNN) imputation when the feature space is not properly scaled?

Accepted Answer

KNN imputation relies on distance metrics (like Euclidean distance). If features are on different scales, a feature with a large range will exert disproportionate influence on the distance calculation, causing the 'nearest' neighbors to be selected based on scale rather than actual similarity. Scaling is mandatory to ensure all features contribute equally to the distance metric.