Using a random forest to select important features for regression