Determining the accuracy in supervised fuzzy classification problems

Abstract

A large number of accuracy measures for image classification are actually available in the literature for cris classification. Overall accuracy, producer accuracy, user accuracy, kappa index and tau value are some examples. But in contrast to this effort in measuring the accuracy in a crisp framework, few proposals can be found in order to determine accuracy for soft classifiers. In this paper we define some accuracy measures for soft classification that extend some classical accuracy measures for crisp classifiers. This elms of measures takes into account the preferences of the decision maker in order to differentiate some errors that in practice may not be have same relevance.