Using multiple and negative target rules to make classifiers more understandable