topic Re: PROC CLUSTER and variable set as a centroid in Statistical Procedures

PROC CLUSTER and variable set as a centroid

ANKH1 — Fri, 11 Oct 2019 19:01:53 GMT

Hi,

We want to identify clusters within a data set based on one variable, use this variable as the centroid of each cluster. Is this possible?

Thanks,

Re: PROC CLUSTER and variable set as a centroid

Reeza — Fri, 11 Oct 2019 19:53:20 GMT

That's just binning at that point with a single variable.
Do a histogram and decide where the cut points are is equivalent IMO.

Re: PROC CLUSTER and variable set as a centroid

ANKH1 — Fri, 11 Oct 2019 20:00:32 GMT

This is after running the proc cluster? We are new with this analysis.

Re: PROC CLUSTER and variable set as a centroid

Reeza — Fri, 11 Oct 2019 20:05:34 GMT

You have one variable that you ran PROC CLUSTER on? It's really only applicable if you have more than one variable. If you have one variable it becomes a binning problem instead and you can easily visualize it. Clustering using PCA primarily because when you start having multiple dimensions it's not possible to view it anymore so have to trust the computer to make the 'clusters'. With a single variable you can easily visualize it and do your analysis.

Re: PROC CLUSTER and variable set as a centroid

ANKH1 — Fri, 11 Oct 2019 20:09:22 GMT

We are looking at multiple variables. How many dimensions is the limit for proc cluster? Should we run a PCA before?

Re: PROC CLUSTER and variable set as a centroid

Reeza — Fri, 11 Oct 2019 20:14:28 GMT

I'm not aware of a limit to the number of variables, just explaining that if you had 2 or 3 three variables you're better of doing it manually. If you have more then you need to move to different methods, such as cluster analysis.

I can't comment on your methodology beyond that because I have no context to the problem you're trying to solve.

Re: PROC CLUSTER and variable set as a centroid

Ksharp — Sat, 12 Oct 2019 11:34:56 GMT

Try PROC FASTCLUS for k-mean cluster method .