On the Suitability of Combining Feature Selection and Resampling to Manage Data Complexity