Why can feature selection help when the number of variables is very large relative to the number of observations?

A. It guarantees that all discarded variables are causal B. It can reduce overfitting risk and computational burden C. It removes the need for model evaluation D. It increases dimensionality intentionally