An extension of the Q diversity metric from single-label to multi-label and multi-ranking Multiple Classifier Systems for pattern classification

Multiple Classifier Systems can show better performance than a single classifier, provided a careful choice of the individual classifiers composing the ensemble. Furthermore diversity among single classifiers, measured through some diversity metrics, is known to be a necessary condition for improvement in the ensemble performance. In this paper we extend the use of the Q diversity metric, a metric used for an oracle output context, to a soft output context for the choice of the best classifier ensemble. We present the Q^t diversity metric, i.e., an extension of the Q metric to multi-label and multi-ranking Multiple Classifier Systems. This metric is tested in a text categorization case study, using the standard Reuters-21578 document corpus and the results strengthen its use in multi-label and multi-ranking Multiple Classifier Systems.