Improved SOM Labeling Methodology for Data Mining Applications

Self-Organizing Maps (SOMs) have been useful in gaining insights about the information content of large volumes of data in various data mining applications. As a special form of neural networks, they have been attractive as a data mining tool because they are able to extract information from data even with very little user-intervention. However, although learning in self-organizing maps is considered unsupervised because training patterns do not need desired output information to be supplied by the user, a trained SOM often has to be labeled prior to use in many real-world applications. Unfortunately, this labeling phase is usually supervised as patterns need accompanying output information that have to be supplied by the user. Because labeled patterns are not always available or may not even be possible to construct, the supervised nature of the labeling phase restricts the deployment of SOM to a wider range of potential data mining applications. This work proposes a methodical and semi-automatic SOM labeling procedure that does not require a set of labeled patterns. Instead, nodes in the trained map are clustered and subsets of training patterns associated to each of the clustered nodes are identified. Salient dimensions per node cluster, that constitute the basis for labeling each node in the map, are then identified. The effectiveness of the method is demonstrated on a data mining application involving customer-profiling based on an international market segmentation study.

Key words: self-organizing maps, neural networks, classification, clustering