An Analysis of Applicability of Genetic Algorithms for Selecting Attributes and Examples for the Nearest Neighbour Classifier

The paper discusses the problem of applicability of genetic algorithms for selecting examples and attributes for the Nearest Neighbour classifier. The emphasis is on the importance of the proper choice of the fitness function and the method for constructing new generations. Some improvements to the existing algorithms are proposed and their fruitfulness is experimentally proved on sixteen benchmark datasets. The empirical evaluation has shown that the increase in the classification accuracy of the modified algorithm is statistically significant.