Abstract
Construction methods of the multiple classification systems using binary classifiers are discussed and evaluated by the system evaluation model based on rate-distortion functions. Suppose the multiclass problems constructed by M (M≧3) categories and N (N≧M-1) discriminators, then they can be solved by the matrices W , where the matrices W are given by the table of M code words with length N. Applying the bench-mark test data (Newspaper articles of 2000 Yomiuri Shinbun), the relationships between the classification error Pe and the number of the binary discriminator N for a given M are investigated, and the systems have desirable properties such as “Flexible", “Elastic", and so on.