Considering Spatialization Aspects When Positioning Input Vectors on Self-Organizing Map Output Grids

Several techniques have been put forward to describe characteristics of a Self-Organizing Map by depicting them on its output grid. These techniques form artificial landscapes, which are also called spatializations. Until now, relatively few methods exist for displaying distinct input vectors on the output grid. Those who exist either do not show patterns in the distribution of the input vectors or do not observe the spatial metaphors implied by the spatializations. This paper proposes an attempt to fill this gap. An approach is introduced where the input vector placement can be influenced by two parameters. The placement technique is tested with two data sets and analyzed through visual inspection. The results show that the approach can both indicate patterns in the input data as well as observe the spatial metaphors of the spatializations. It thereby allows for a meaningful combination of these visualization forms.