توضیحات
ABSTRACT
Despite the sizable achievements obtained, the use of soft classi®ers is still limited by the lack of well-assessed and adequate methods for evaluating the accuracy of their outputs. This paper proposes a new method that uses the fuzzy set theory to extend the applicability of the traditional error matrix method to the evaluation of soft classi®ers. It is designed to cope with those situations in which classi®cation and/or reference data are expressed in multimembership form and the grades of membership represent dierent levels of approximation to intrinsically vague classes. Ó 1999 Elsevier Science B.V. All rights reserved.
INTRODUCTION
In many applications, it is desirable to have a “soft” classi®er that, for a given input pattern vector, computes the “likelihood” that the pattern lies in any of a set of possible classes. In general, soft models for classi®cation are rooted in speci®c representation frameworks within which the partial belongingness of a given pattern to several categories at the same time is explicitly modeled (Binaghi et al., 1996; Bouchon- Meunier et al., 1995). Statistical classi®cation models interpret a given pattern as fully contributing to a given class, and the computed probabilities are an expression of the frequency with which this full membership occurs. In soft models for classi®cation, non-probabilistic uncertainty due to vagueness and/or ambi guity should be modeled as partial belongingness to several categories at the same time (Klir and Folger, 1988). Various approaches may be used to derive a soft classi®er. These approaches are based on speci®c uncertainty representation frameworks such as the fuzzy set theory, Dempster±Shafer theory and certainty factors (Binaghi et al., 1996; Bloch, 1996). In addition to the use of speci®c representation frameworks, the output of “hard” classi®ers, such as the maximum likelihood classi®er and the multilayer perceptron, can be softened to derive measures of the strength of class membership(Schowengerdt, 1996; Wilkinson, 1996) . The most common solutions adopt a fuzzy set framework (Pedrycz, 1990; Binaghi and Rampini, 1993; Ishibuchi et al., 1993). The apparatus of the fuzzy set theory serves as a natural framework for modeling the gradual transition from membership to non-membership in intrinsically vague classes. Here the assumption is that the classi®cation process is possibilistic in nature.
Year: 1999
Publisher : ELSEVIER
By : Elisabetta Binaghi a , Pietro A. Brivio b, Paolo Ghezzi b, Anna Rampini a
File Information: English Language/ 14 Page / size:674 KB
Download: click
سال : 1999
ناشر : ELSEVIER
کاری از : Elisabetta Binaghi a, Pietro A. Brivio b, Paolo Ghezzi b, Anna Rampini a
اطلاعات فایل : زبان انگلیسی / 14 صفحه / حجم : KB 674
لینک دانلود : روی همین لینک کلیک کنید
نقد و بررسیها
هنوز بررسیای ثبت نشده است.