Organizing and Exploring High-Dimensional Data with the Growing Hierarchical Self-Organizing Map

The self-organizing map is a very popular unsupervised neural network <br> model for the analysis of high-dimensional input data as in data mining <br> applications.<br> However, at least two limitations have to be noted, which are caused, on <br> the one hand, by the static architecture of this model, as well as, on the <br> other hand, by the limited capabilities for the representation of <br> hierarchical relations of the data.<br> With our growing hierarchical self-organizing map we present an artificial <br> neural network model with hierarchical architecture composed of <br> independent growing self-organizing maps to address both limitations.<br> The motivation is to provide a model that adapts its architecture during <br> its unsupervised training process according to the particular requirements <br> of the input data. The benefits of this neural network are first, a <br> problem-dependent architecture, and second, the intuitive representation <br> of hierarchical relations in the data. This is especially appealing in <br> exploratory data mining applications, allowing the inherent structure of <br> the data to unfold in a highly intuitive fashion. <br> <br>