Developing parsimonious ensembles using ensemble diversity within a reinforcement learning framework