| 摘要 |
PROBLEM TO BE SOLVED: To appropriately classify special statistical numeric data such as medical data, and to make a potential Dirichlet distribution method applicable to even contiguous numeric data depending on the classified result.SOLUTION: A frequency classification unit 2 performs classification based on deviation in frequency on numeric data of a prescribed item obtained from a sample set, and a numeric classification unit 3 performs classification based on deviation in numeric. First, inputted numeric data is classified, by frequency deviation, into regular data and upper-end and/or lower-end non-regular data, and classification by deviation in numeric and classification by deviation in frequency are applied recursively to the non-regular data and the regular data, respectively, which continues until a separation result converges. An estimation unit 5 assigns a numeric label to each of the separation results, generates a document based on the label in the item data of each sample, and applies a potential Dirichlet distribution method to a document set covering all samples, thereby an analytical result for the initial continuous numeric data is obtained. |
