Abstract
Z. Pawlak has proposed a new concept of approximate data analysis based on rough sets and applied it to medical analysis. Rough sets are defined by equivalence relations and the data base is called an information system. A method of reducing attributes in the given information system has been developed by equivalence relations with regard to attributes.
In this paper, we propose a new method of reducing information systems by considering the classification given by experts, although Z. Pawlak's method does not consider the given classification in reducing information systems. Thus our method can reduce more attributes than Z. Pawlak's method. Also, a method of constructing a fuzzy expert system is described by the proposed method of reducing information systems and by introducing fuzzy intervals represented as fuzzification of data values.
As an example, the fuzzy expert system for medical diagnosis is built by fuzzy inference rules. There are 367 fuzzy if-then rules constructed by the lower approximations which show consistency of the given data and the classification given by experts. It is very useful that inconsistency of test data and expert's diagnoses can be clarified by our approach.
