There are eight unique features that differentiate FocalPoint Clustering from other k-means analysis programs: 1. Tests case order sensitivity in trials 1. Clustering procedures are notoriously sensitive to the case order, and k-means analysis is no exception. Whereas other programs produce only one final cluster solution, FocalPoint tests the sensitivity of different case orders and compiles a range of cluster solutions that can then be compared and evaluated.
2. Because FocalPoint uses an exact relocation test on the Euclidean Sum of Squares, convergence is assured and goodness-of-fit is calculated.
3. FocalPoint finds several "top solutions" and lets you decide which of these to progress for action as a cluster model.
4. There are six ways of specifying the start of a FocalPoint cluster analysis. You can section a tree, extract cliques or select exemplars from it; choose seed points from another cluster solution; randomly assign clusters; or specify a segmentation.
5. FocalPoint calculates the reproducibility of each cluster solution from its frequency of occurrence over a large number of random trials.
6. Sub-optimal, solutions often exhibit small or singleton clusters. These "outliers" are remote from the densest area of data. FocalPoint allows such outliers to be identified and removed from the cluster solution, so that they do not distort the cluster means.
7. Conventional k-means analysis forms clusters around means. You can do this with FocalPoint or, alternatively, cluster around "exemplars" - the most typical cases in each cluster. The advantage of clustering around exemplars is that the cluster centres are actual cases.
8. When you have a cluster solution that is actionable, FocalPoint provides guidance on the important variables that discriminate between the clusters. You can use this to revise the variable weights, thereby calibrating your cluster model to your key variables and reducing the effect of noise variables.
Full and trial versions of FocalPoint are available. To find out more, |