The data classifier, data taxonomic method and records the program for data classification the record medium

摘要

PROBLEM TO BE SOLVED: To provide a high performance data classifying device suitable for attribute data used for credit giving, marketing, etc. SOLUTION: An attribute item unifying part 31 decides which the attribute item of an attribute category is unified to between a category type and a numerical type. When the attribute item is unified to the category type, a data accumulating part 32 performs cross tabulation of the number of data corresponding to each category. A number of input attribute items deciding part 33 decides the number of attribute items to be inputted to a neural network. An input data shaping part 34 makes data into a shape fitting an input of a hierarchical neural network. A BP learning part 35 decides the structure of the neural network by using results of the part 33 and performs learning by a BP method by using learning data of the part 34. A BP estimating part 36 performs classification estimation of new attribute data by using BP learning results and new attribute data. Thus, it is possible to obtain stable and high classification performance because of using a number of intermediate layer units deciding method corresponding to the quality of attribute data.