Fuzzy c-means -- initialisation?

Jonathan G Campbell (jg.campbell@ulst.ac.uk)
Mon, 20 Apr 1998 14:48:37 +0200 (MET DST)

I'm implementing a fuzzy c-means clustering routine -- following p.360
of G.J. Klir and B. Yuan, 1995, Fuzzy Sets and Fuzzy Logic, Prentice
Hall.

Is there any agreed `best' method of initialisation of the memberships?
-- presumably best in the sense of faster convergence, I assume that
local optima is not a problem.

At present, I intend initialising as per my k-means clustering routine,
i.e. randomly allocate class labels; for fuzzy version, I'll simply
initialise memberships to {0.0, 1.0}.

Many thanks in anticipation.

Jon Campbell

-- 
Jonathan G Campbell Univ. Ulster Magee College Derry BT48 7JL N. Ireland 
+44 1504 375367 JG.Campbell@ulst.ac.uk  http://www.infm.ulst.ac.uk/~jgc/