Learning metrics for self-organizing maps
Samuel Kaski, Janne Sinkkonen, Jaakko Peltonen · 2002
We introduce methods that adapt the metric of the data space to reflect relevance, as indicated by a auxiliary data associated with the primary data samples. The derived metric is especially useful in descriptive data analysis by unsupervised methods such as the self-organizing maps. In this work we use the new metric to refine SOM-based analyses of the factors affecting the bankruptcy risk of companies.