Visualizing non-hierarchical and hierarchical cluster analysis with clustergrams (Q1887244)

A new graphical tool for cluster analysis, named clustergram, is described. It can be used for the visualization of consecutive hierarchical or non-hierarchical clusterizations with growing number of clusters (e.g., 1 through 10). The clusters are represented by the mean values of some characteristics of the classified objects. The cluster means for consecutive clusterizations are connected by lines (parallelograms) with width proportional to the number of objects which flowed from the preceding cluster to the following. This method is compared to the dendrograms for hierarchical clusterization. An analysis of data on asbestos claims and Thailand landmine data is presented.