Categorical classifiers in multiclass classification with imbalanced datasets

This paper discusses, in a multiclass classification setting, the issue of choice so-called categorical classifier, which is procedure or criterion that transforms probabilities produced by probabilistic classifier into single category class. The standard Bayes Classifier (BC), but it has some limits with rare classes. studies performance BC versus two alternatives, are Max Difference (MDC) and Ratio (MRC), through an extensive simulation case studies. results show both MDC MRC preferable to setting imbalanced data.