A classification method based on the nominal attributes quantization

There are many nominal attributes in the decision table, and their values only stand for the conception partition but not the meaning the values have. So some statistical analysis methods that have a high classification precision but need numerical values cannot be used. The paper introduces a method that converts nominal attributes to numerical attributes at first, and then uses the Fisher discriminate method in multistatistics analysis to build the discriminant function of classification. Besides it, in order to deal with nonlinear datum, we also cite the kernel Fisher discriminant to further improve the classification precision. At last, experimental result confirms its effectiveness.