An arbiter (Chan & Stolfo, 1995) is a separate, single classifier, which is
trained on a subset of the original data. This subset consists of examples on
which the base classifiers disagree. They also investigate arbiter trees, in which
arbiters that specialize in resolving conflicts between pairs of classifiers are organized
in a binary decision tree. Arbiters use information about the disagreement
of classifiers for selecting a training set.