On reducing information systems by accuracy measures in rough sets

In this paper, we discuss on reducing information systems from some aspects. Assuming that some classes are more important than the other in medical diagnosis, all the classes are divided into two groups by considering importance grades of classes. In reducing attributes, the accuracy measures of the important group should not be reduced, while those of the other group can be slightly reduced. Furthermore, introducing an order relation with regard to importance grades, the given information system is reformed so that predictive rules from reformed one are permissive for patients in medical diagnosis.