1) For an unsupervised classification application,we expect
results like the ones shown in Fig. l(a), where we obtained
three weight vectors located at the mean vectors of the three
classes or clusters. In this case, the recognition rate will be
the highest when all the samples are classified by distance
classifiers based on the three weight vectors. However, in the
case shown in Fig. l(c), the recognition rate will significantly
decrease since there are two disturbing units located at the
boundary points of different clusters and the two units will
draw quite a large portion of samples from three of the four
clusters to form two mixture groups which give an incorrect
clustering result.