How to handle confound variables?
A r/MachineLearning post discusses confounding by range in automotive radar point-cloud classification; the poster worries models learn distance instead of class and asks about strategies like dataset splitting or removing the feature.