TABLE OF CONTENTS This paper presents a new method for cluster analysis based on unidimensional scaling (UDS). UDS is a technique to find the coordinates of n objects on a real line so that the interpoint distances can best approximate the observed dissimilarities between pairs of objects. First, we propose a simple and effective way to find the coordinates of the objects by minimizing the squared error fucntion. Then the coordinates of these n objects are used to detect the hidden clusters. Large jumps or gaps in the coordinate indicate possible boundaries of two clusters. Real and simulated examples are used to illustrate and to compare our method with the K-Means clustering.
Cluster analysis is a procedure to find hidden groups in data. Many clustering algorithms are available and they are divided into hierarchical or nonhierarchical clustering methods. Among the nonhierarchical clustering methods, K-means clustering is most common and widely used method since it can classify large number of objects easily. However, there are some problems with this K-means clustering. For example, the number of clusters has to be specified in advance, the result depends on the initial seed.
In this paper, we present a new nonhierarchical clustering method based on unidimensional scaling(UDS). The main purpose of UDS is to find the...
