580 California St., Suite 400
San Francisco, CA, 94104
Figure 1 – uploaded by Witold Pedrycz
Figure 1 One can note that the classification boundaries become more complicated once we move apart from the core portion of the clusters. The cores themselves come with a rectangular format of the class boundaries that fully complies with the type of distance being considered there. From the classification standpoint, we may say that the core part of data requires a simple type of a classifier whereas to deal with a residual section of patterns we require a classifier of substantially higher level of complexity. Evidently, most of classification error comes associated with the residual structure of the patterns.