Finnish car rejection reasons shown in an interactive SOM visualization tool

Abstract

In this paper a new SOM visualization tool is introduced. It is shown how Collaborative Filtering can be used as preprocessing before the SOM training. Our goal was to provide for the user a possibility to analyze car differences by component planes which is not possible by original published tables. In addition it is possible to explore how different flaws are related in time or with other variables. The effects of the driver dependent components, such as tires, can be filtered out from rejection probability using the component plane codebooks. The interactive SOM visualization is very useful when a large number of labels is present. We developed a function to generate the needed files for a Processing language based tool. Our tool can be used simultaneously with the SOM Toolbox.