An extended self-organizing map (ESOM) for hierarchical clustering

The bottom-up hierarchical clustering methodology that is introduced in this paper is an extension of self-organizing map neural network (ESOM) and it provides remedy for two different major problems. The first one is related to the hierarchical clustering and the second one is related to the self-organizing map (SOM) neural network that is able to perform a clustering task. The crucial problem that the hierarchical clustering approaches (top-down and bottom-up) are faced with is the fact that once a merging or decomposing of two clusters takes place, it is impossible to undo or redo it. The crucial problem for SOM stems from the fact that the initial clusters´ weight vectors, that are generated randomly, highly influence the outcome of the SOM clustering.