shows that the four initialization methods almost always make the K-Means algorithm able to reach what is supposed to be the best final partition.
Best square-error values over 10 runnings of the GA per entry, i.e., left extremes of the probability distributions (best instance order)