Genetic Algorithm Based k Nearest Neighbors

Abstract We develop k Nearest Neighbor (kNN) techniques using genetic algorithms and decision trees: decision trees providing problem-specific infonnation able to adjust the kNN parameters, and genetic algorithms to optimize them further. Their combined use shows to be quite effective. Indeed, the problem-specific infonnation provided by decision trees benefits enonnously the search of genetic algorithms: it reduces dramatically their computing effort while enhancing their optimization capabilities. The resulting methodology is quite general. In this paper it is developed, tuned and illustrated in the particular context of power system transient stability.