Hi, I have a very large dataset of 3008 individuals and 800 numerical variables. In fact it is a table of 3008 36-monthes multivariated time series that I would like to classify with an unsupervised algorithm I had a look at the function kkmeans of e1071 package, which seems to be a kernel weighted version of the algorithm algorithm, and the bclust from the same package which does bootstrapping of kmean algorithm (bagging) I had the idea of combining these two functions, in fact using kkmeans in bclust parameters. What should I do ? Reducing dimension and then classify, or classifying only ? Should I use kkmeans only ? or bclust only ? I'm very interested in your opinions. Laurent. -- «l'homme chéri ne dure plus d'un jour» «Moi» @<http://www.le-valdo.com> [[alternative HTML version deleted]]