On the use of rule-sharing in learning classifier system ensembles