Decision-theoretic approaches in fuzzy rule generation for diagnosis and fault detection problems.
Sebastian Beck, Ralf Mikut, Jens Jäkel, Georg Bretthauer · 2003
A typical task in technical fault detection or medical diagnosis problems is to discriminate normal behavior from one or more types of abnormal behavior by means of different measured or computed features. This may lead to difficult classification problems due to extremely different a priori probabilities of classes and heterogeneous classes (e. g. unknown sub-classes for different errors to be detected). In this paper, an approach to design fuzzy classifiers is presented, which is based on decision-theoretic measures and uses a learning data set with feature values and given information about decision and classifier costs.