Computing theoretically-sound upper bounds to expected support for frequent pattern mining problems over uncertain big data