which method is useful for dimensionality reduction especially for small data-sets?