A comparison of map neuron labeling approaches for unsupervised self-organizing feature maps

Abstract

The self-organizing map (SOM) is an unsupervised neural network approach that reduces a high-dimensional data set to a representative and compact two-dimensional grid. In so doing, a SOM reveals emergent clusters within the data. Research has shown that SOMs lend themselves to visual and computational analysis for exploratory and data mining purposes. However, an important requirement for many SOM interpretations is the characterization of the mappsilas emergent clusters. This process is often addressed by either a manual or automated map neuron labeling approach. This paper discusses techniques for the labeling of the unsupervised, supervised and semi-supervised variants of the SOM, and proposes some new methods. It also presents empirical results characterizing the performance of two automated labeling approaches for fully unsupervised SOMs when applied for example classification of experimental data sets.