A kNN method that uses a non-natural evolutionary algorithm for component selection

Abstract

This paper details an evolutionary algorithm that forms a new population by combining genes of three members of the current population. The first member is the best member of the population, the second one is the current member to be replaced and the third one is a member chosen randomly from the current population. We used this algorithm for component selection of a kNN (k Nearest Neighbor) method for breast cancer prognosis. Results with the UCI prognosis data set show that we can find components that help improve the accuracy of kNN by almost 3%, raising it above 79%.