Classification of Chain-Link and Other Data with Spherical SOM

A new classification method is proposed based on the spherical SOM that has been developed earlier for visualizing multidimensional data sets. Phase distances between labeled data on the spherical surface are computed. With these distances, a dendrogram can be constructed. Then, using the constructed dendrogram, a classification of each cluster group on the spherical surface, based on the label data was carried out. This method can be applied to various data sets. Here, the method was applied to the chain-link problem which can be considered as a particularly difficult one from a representational standpoint, and to the problem of separating parallel random number planes.