Visualising Clusters in Self-Organising Maps with Minimum Spanning Trees

The Self-Organising Map (SOM) is a well-known neuralnetwork model that has successfully been used as a data analysis tool in many different domains. The SOM provides a topology-preserving mapping from a high-dimensional input space to a lower-dimensional output space, a convenient interface to the data. However, the real power of this model can only be utilised with sophisticated visualisations that provide a powerful tool-set for exploring and understanding the characteristics of the underlying data. We thus present a novel visualisation technique that is able to illustrate the structure inherent in the data. The method builds on minimum spanning trees as a graph of similar data items, which is subsequently visualised on top of the SOM grid.