Using fuzzy methods to model nearest neighbor rules

The basic principle used in the construction of nearest-neighbor models is discussed. The induced ordered weighted averaging (IOWA) operators are shown to provide a useful formal structure for building nearest-neighbor models. A methodology for learning IOWA operator nearest-neighbor models is described. Various types of nearest-neighbor rules are investigated, including those based on a linguistic specification. The situation in which the value of interest lies in an ordinal set is also considered. It is shown that the weighted median provides a useful tool for constructing nearest-neighbor rules in this case