An empirical investigation on the use of diversity for creation of classifier ensembles