Skip to content
Prev 199222 / 398506 Next

Non parametric clustering

Dear Giorgio,

Fixed Point Clustering is somewhat related to the mean shift algorithm, 
see fixmahal in the package fpc and the references given there (note that 
this is for overlapping clustering, not partitioning).
Another potentially useful method could be dbscan, as well implemented in 
fpc.
mclust offers Gaussian mixtures with a uniform "noise" component (may make 
sense after transforming your variables).
trimcluster has trimmed k-means, if you look for spherical clusters and 
the problem with normality is basically outliers or heavy tails.
You may also have a look at pam/clara in package cluster.

Best regards,
Christian
On Thu, 5 Nov 2009, giorgio.arcara at unipd.it wrote:

            
*** --- ***
Christian Hennig
University College London, Department of Statistical Science
Gower St., London WC1E 6BT, phone +44 207 679 1698
chrish at stats.ucl.ac.uk, www.homepages.ucl.ac.uk/~ucakche