Classification with multiple prototypes

We compare learning vector quantization, fuzzy learning vector quantization, and a deterministic scheme called the dog-rabbit (DR) model for generation of multiple prototypes from labeled data for classifier design. We also compare these three models to three other methods: a dumping method due to Chang (1974); our modification of Chang´s method; and a derivative of the batch fuzzy c-means algorithm due to Yen-Chang (1994). All six methods are superior to the labeled subsample means, which yield 11 errors with 3 prototypes. Our modified Chang´s method is, for the Iris data used in this study, the best of the six schemes in one sense; it finds 11 prototypes that yield a resubstitution error rate of 0. In a different sense, the DR method is best, yielding a classifier that commits only 3 errors with 5 prototypes