Neighborhood size selection in the k-nearest-neighbor rule using statistical confidence