Divisive cluster analysis
|
|
Top Previous Next |
|
In divisive clustering methods, the samples (columns) all start as members of a single group which is subsequently subdivided. TWINSPAN is an example of such an approach. The method available under Clustering: Divisive minimises the sums of squares of the dissimilarity of each cluster formed. To carry out the analysis, you must specify the number of clusters required in the Setup dialog window. The number of clusters must lie between 1 and the number of samples present in your data set; the setup dialog will inform you how many samples there are (see image below).
To gain an idea for a suitable number of clusters you might find it useful to examine a PCA or MDS plot. Although the method will always group the samples into the number of clusters you request, the results will not necessarily be meaningful for every data set, and so the method is best used when some other analysis such as TWINSPAN or PCA has suggested the presence of a number of groups, and you wish to see if an independent method can produce the same group membership.
|
