How k-nearest neighbors algorithms classify data