Automatic finding of good classifiers following a biologically inspired metaphor

摘要: The design of nearest neighbour classifiers can be seen as the partitioning whole domain in different regions that directly mapped to a~class. definition limits these is goal any based algorithm. These described by location and class a reduced set prototypes rule. rule defined distance metric, while matter design. To compute this prototypes, most algorithms literature require some crucial parameters number use, smoothing parameter. In work, an evolutionary approach on Nearest Neighbour Classifiers (ENNC) introduced where no are involved, thus overcoming all problems derived from use above mentioned parameters. algorithm follows biological metaphor each prototype identified with animal, territory animals. animals evolve competitive environment limited resources, emerging population able survive environment, i.e.~emerging a~right for classification objectives. has been tested using domains, showing successful results, both accuracy distribution achieved.