A fuzzy neural network approach based on Dirichlet tesselations for nearest neighbor classification of patterns

A neural network classifier using fuzzy set representation of pattern classes is presented. Network construction and learning is performed incrementally in a single pass by building an aggregate of space-filling regions that constitutes a simplified variant of the construction known as Dirichlet tesselation (or Voronoi diagram). Each region is delimited by a set of hyperplanes and is endowed by a fuzzy membership function that forms the basis of learning and recall. Experimental results concerning difficult recognition problems show that the proposed approach is very successful in applying fuzzy sets to pattern classification