Fitness functions in editing k-NN reference set by genetic algorithms

摘要

In a previous paper the use of GAs as an editing technique for the k-nearest neighbor (k-NN) classification technique has been suggested. Here we are looking at different fitness functions. An experimental study with the IRIS data set and with a medical data set has been carried out. Best results (smallest subsets with highest test classification accuracy) have been obtained by including in the fitness function a penalizing term accounting for the cardinality of the reference set.