| Popis: |
There have been many alternative ways in which clusters have been defined. Perhaps the most frequent choice has been the geometric concept of a cluster as a set of point ‘close’ in some space, a concept related to notions of probability density functions and hence to the framework of mathematical statistics. However such a model is not everywhere suitable, and in this paper I shall also examine some of the alternatives, chosen from models which have been used in deciding the number of clusters present. The aim of this examination it twofold. Firstly to indicate what alternatives have in fact been suggested, for many of them are neither well-known nor widely applied. Secondly to try and explore the situations in which one definition might be more appropriate than another. Ultimately such a decision must rest with the analyst, or agent, for approaches to testing for existence of clusters, and for determining the number of clusters, are closely related to the nature of the clusters being sought. |
