A neutral zone classifier for three classes with an application to text mining

Abstract

A classifier may be limited by its conditional misclassification rates more than its overall misclassification rate. In the case that one or more of the conditional misclassification rates are high, a neutral zone may be introduced to decrease and possibly balance the misclassification rates. In this paper, a neutral zone is incorporated into a three-class classifier with its region determined by controlling conditional misclassification rates. The neutral zone classifier is illustrated with a text mining application that classifies written comments associated with student evaluations of teaching.