From: earlcox@earlcoxreports.com   
      
   Try connecting fuzzy c-means or the adaptive fuzzy cluster technique to a   
   genetic algorithm that breeds a random number of cluster centers and then   
   measures the degree of cluster compactness (remember that in fuzzy   
   clustering a data point can belong to multiple clusters (with different   
   degrees of membership) so some component of the fitness function should   
   attempt to minimize the number of overlapping fuzzy sets thus partitioning   
   the data points into the minimum number of shared fuzzy regions). . In work   
   I've done in managed healthcare fraud detection, project risk assessment,   
   and customer segmentation this has worked extremely well. You can then   
   convert clusters to rules by treating the cluster centers as a fuzzy number   
   by applying one of the approximation hedges to the centroid and to each of   
   the dimensions, thus, suppose there are four dimensions in the data, a, b,   
   c, d and this has five clusters c1(a,b,c,d), c2(a,b,c,d), etc. then a rule   
   for the first cluster with incoming date point (x1,x2,c3) might be,   
      
   if x1 is around(c1(a)) and x2 is around (c1(b)) and x3 is around(c1(c)) then   
   y = around(c1(d))   
      
   since clustering is an unsupervised knowledge discovery approach, we are   
   free to consider any of the dimensions as the dependent variable and treat   
   the remaining dimensions as the independent variables.  The expectancy   
   (width of the fuzzy number) for the each of the around hedge operators can   
   be determined by computing the contingent density of the cluster points when   
   sliced at the associated dimension (that is, we compute something like the   
   standard deviation of the clustering to determine whether the kurtosis of   
   the distribution is leptokurtic, mesokurtic, or platykurtic and use one half   
   this distribution statistic as the radius of the bell curve).   
      
   Earl   
      
      
   Earl Cox   
   Founder and President   
   Scianta Intelligence, LLC   
   Turn Knowledge Into Intelligence   
      
      
   AUTHOR:   
   "The Fuzzy Systems Handbook" (1994)   
   "Fuzzy Logic for Business and Industry" (1995)   
   "Beyond Humanity: CyberEvolution and Future Minds"   
   (1996, with Greg Paul, Paleontologist/Artist)   
   "The Fuzzy Systems Handbook, 2nd Ed." (1998)   
   "Fuzzy and Evolutionary Strategies in Data Exploration and Modeling"   
   (due Early Fall 2004)   
      
   "Gysber J. TAMAELA"  wrote in message   
   news:881ead09.0407112246.6f05c444@posting.google.com...   
   > Dear all,   
   > I have several questions.   
   > 1. Besides Fuzzy Substractive Clustering, is there any fuzzy   
   > clustering method, which undefined the number of cluster?   
   > 2. What is the non fuzzy clustering that can be camparable with fuzzy   
   > one?   
   > 3. What, why, when, where and how, Fuzzy association rule is?   
   >   
   > Thank you.   
   >   
   > ps: to Mr. Stiller, thank you for the URL!   
   >   
   > regards,   
   >   
   > Gysber J. TAMAELA   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)