On semi-supervised fuzzy c-means clustering