The self-organizing map (SOM) forms a nonlinear projection from a high-dimensional data onto a regular (usually) two-dimensional grid and have been widely studied as a software tool for visualization of high-dimensional data. However, the automation of knowledge discovery in SOM is not straightforward. A complementary stage is need in trained SOMs when dealing with complex data clustering problems. This paper investigates the usage of contiguity constrained hierarchical clustering in SOM

more »

... . The objective is to find regions of neurons that are related classes of input data. The distributed prototype representation enables the discovery of complex geometries and even nonlinear separable data clusters in an unsupervised way. It is also shown that the algorithm can also estimate the probable number of clusters.