A new learning approach based on cooperative coevolution is proposed for
   classification problems with multiple outcomes. The number of
   populations that coevolve equals the number of outcomes of the
   classification problem. Each population evolves rules with the same
   outcome; when the quality of an individual (rule) is computed,
   collaborators from all others populations are selected in order
   to form a complete rule set to be applied to the training data.
   In the end of the evolutionary process, a set of rules is extracted from all
   final populations applied to the test data. Experiments
   were performed on a well-known benchmark problem and results encourage
   further investigation.