APPARATUS, METHOD AND PROGRAM FOR CLUSTER ANALYSIS USING K-MEANS METHOD, AND RECORDING MEDIUM RECORDING THE SAME PROGRAM

摘要

PROBLEM TO BE SOLVED: To provide a cluster analysis apparatus using a k-means method capable of determining a proper number of clusters and an initial center position, using a lower limit value of the number of cluster elements. SOLUTION: The method includes a means (1) for rearranging data in either an ascending order or a descending order; a means (2) for calculating an interelement distance of the data after the rearrangement; a means (3) for determining a division width from the interelement distance; a means (4) for dividing an analysis space into small grid spaces using the division width; a means (6) for calculating a cluster initial center position, using the greatest small grid space having the greatest number of elements and surrounding small grid spaces; and a means (7) for performing the k-means method using the cluster initial center position. The division means (4) selects a tentative node point data having a greater interelement distance than a threshold, and selects as a node point data a tentative node point data having data of which number is greater than a predetermined number existent between adjacent tentative node point data, and determines, as a division width, the smallest difference value between the adjacent node point data. COPYRIGHT: (C)2005,JPO&NCIPI