I've items that have a geo-spatial position and a temporal origin. For both dimensions, I build clusters so far.

I'm now in search of a way to merge this different clusters forming spatio-temporal clusters. Of course, I want to prevent calculating completely new clusters from the scratch and rather use the existing information through the previous clustering.

Is there any algorithm how to build 3d clusters by merging a previous 1d and a 2d clustering process?


  • 1
    $\begingroup$ I would speculate that, in restricting the algorithm to combining clusters you may get a poorer result than if you started from scratch, because you disallow "interactions" between the geo-spatial and temporal effects. $\endgroup$ Jan 25, 2011 at 6:14
  • $\begingroup$ What is your data? What do your space-time observations represent? $\endgroup$
    – b_dev
    May 25, 2011 at 22:42
  • $\begingroup$ Documents that have a time and location of origin. $\endgroup$
    – b_erb
    May 26, 2011 at 7:33

1 Answer 1


I don’t know if I understood your question correctly, but I’ll give it a try. Any hierarchical agglomerative algorithm can do the job. Remember that agglomerative algorithms proceed by “pasting” observations to a cluster and treating the cluster as a single unit.

I would suggest this:

  1. Substitute your 1d or 2d observations with the clusters’ centroids.

  2. Use a distance to assign the temporal observations into the 1d/2d cluster. (Euclidean distance might work).

Hope this works =)

  • $\begingroup$ Byt the way, for more information about hierarchical agglomerative clustering: Izenman, Modern Multivariate Statistical Techniques. Springer. 2008. $\endgroup$
    – deps_stats
    Jan 25, 2011 at 16:19

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.