Managing Category Proliferation in Fuzzy ARTMAP Caused by Overlapping Classes

This paper addresses the difficulties brought about by overlapping classes in fuzzy ARTMAP (FAM). Training with such data leads to category proliferation, and classification is made difficult not only by the large number of categories but also the fact that such data can belong to either class. In this paper, changes were proposed to allow more than one class to be predicted during classification, and a number of modifications were explored to reduce the number of categories. The excessive creation of small categories was suppressed with the implementation of the modifications, and the predictive accuracy improved despite the significant reduction in number of categories. No major changes needed to be made to the FAM architecture.