Assessing Rough Classifiers

Article type: Research Article

Authors: Peters, Georg

Affiliations: Munich University of Applied Sciences, Munich, Germany & Australian Catholic University, Australia. georg.peters@cs.hm.edu

Note: [] Address for correspondence: Department of Computer Science and Mathematics, Lothstrasse 34, 80335 Munich, Germany

Abstract: Since its introduction a prime area of application of rough sets theory has been in the field of classification. In this area rough sets theory provides a powerful toolbox of methods to deal with incomplete and contradicting information. Obviously, the assessment of the obtained classification results is of crucial importance. In our paper, we propose and evaluate some rough performance indices to evaluated the quality of bi- and multinomial classifiers. To illustrate their characteristics we perform comparative experiments on a synthetically generated data set.

Keywords: Rough Sets, Classification, Performance Indices, Boundary Roughness

DOI: 10.3233/FI-2015-1191

Journal: Fundamenta Informaticae, vol. 137, no. 4, pp. 493-515, 2015