topic Re: Cluster Analysis in SAS Data Science

Cluster Analysis

wutao9999 — Wed, 02 Nov 2016 15:39:45 GMT

I have a data set for cluster analysis. The code I used for the analysis is listed as below. My question is that: Assuming there are 100 data points in the data set, how can I define the minimum number of data points is clustered in each group?

Right now, I see some cluster, outputted by the current method, only contains 1 data point in a cluster. This is why I am wondering if I can set a minimum number, such as the number of data points in a cluster must be more than 10% or 20%, etc. for example.

Thank you for the help.

PROC FASTCLUS DATA=model_data3

MAXC=3

MAXITER=100

REPLACE=FULL

OUT=WORK.CLKMKMeansData

;

VAR volume;

RUN;

Re: Cluster Analysis

PGStats — Wed, 02 Nov 2016 19:37:35 GMT

It might be a better idea to increase MAXCLUSTERS and to consider observations which end up alone in a cluster as outliers,

Re: Cluster Analysis

PGStats — Wed, 02 Nov 2016 20:02:46 GMT

Since you have a single clustering variable, you could also try to fit a finite mixture model:

PROC FMM DATA=model_data3;
model volume / kmax=3;
output out=CLKMKMeansData / group=volumeGroup;
run;

(untested)