Self Organizing Maps — Application to Census Data

Abstract

In the previous chapter, Stan Openshaw provides a lucid overview of the classification of spatial data using neural-nets. In this chapter we follow with an example that uses one form of neural classification, namely the Kohonen Self Organizing Map (SOM). In this case, the SOM is used to organize demographic data gathered from census information in order to investigate population groupings and their spatial distribution. In other words, to perform an unsupervised classification, or a mapping of a ten dimensional input to a simple two dimensional surface, or in the full jargon, a nonlinear projection onto two dimensions of the probability density function. As this is an example of the techniques already discussed, we assume that we need not repeat the theory and practice of the preceding chapter.