If that is the case you could write a normal algorithm to get the proportions and see if it separates you data set nicely. (which should be done to prove this assertion)