ABSTRACT
The present chapter makes an argument in favor of understanding and utilizing the notion of causality for feature selection: from an algorithm design perspective, to enhance interpretation, build robustness against violations of the i.i.d. assumption, and increase parsimony of selected feature sets; from the perspective of method characterization, to help uncover superfluous or artifactual selected features, missed features, and features not only predictive but also causally informative.
