The use of a novel genetic algorithm in component selection for a kNN method for breast cancer prognosis

摘要

This paper shows the application of a genetic algorithm (GA) for component selection to improve the accuracy of a kNN (k-Nearest Neighbor) method when using it for breast cancer prognosis. Our GA uses the best chromosome (member) in a generation to produce new ones. The probabilities of crossover and mutation have not fixed values, instead they depend on the evaluation of the chromosomes involved in the generation of new ones. The GA determines the best components that must be used in prognosis by the kNN method. This method for the UCI breast cancer data usually gives a 76% average accuracy, but we have found a combination of only sixteen components that rises the average accuracy of kNN to 79%.