Feature selection is preferred for dimensionality reduction when a dataset contains irrelevant or redundant variables that can simply be discarded. In contrast, feature extraction is used when most of the original information must be retained but explained using fewer, newly constructed dimensions.