one approach of feature selection is filter method.
In this method we check 2 things:
relation between feature i and target. we want to preserve only features that influence the target
relation between feature i and feature j. we want to remove redundant features
There can be a situation where there is no direct relation between feature i to target, but there is relation between combination of 2 or more features to the target.
from my understanding, all the filter methods checks direct relationship (1 or 2) between feature i and target (and they don't check relation between multiple features to target)
Am I right ?
Are there any filter methods to check non-direct relation between features and target ? (or we must use wrapper or hybrid methods to select features when there is no direct relation ) ?