Building models with SAS Enterprise Miner, SAS Factory Miner, SAS Visual Data Mining and Machine Learning or just with programming

Clustering nominal variables in Enterprise miner 14.2

Accepted Solution Solved
Reply
Occasional Contributor
Posts: 12
Accepted Solution

Clustering nominal variables in Enterprise miner 14.2

Hi,

I have a data set with over 30 attributes, mostly nominal, 3 ordinal, 3 binary variables, and just one interval variable (age).  

Could anyone please advise if I can successfully do a clustering on this data set using Enterprise Miner 14.2?

 

  1. Does the Cluster Node support the clustering of data with nominal variables?
  2. Does the Cluster Node support the clustering of data with both numeric and nominal variables?
  3. Can I use the HP Cluster node instead (on a single-machine mode)? How does this differ from using the normal cluster node (on a single-machine mode)?
  4. If EM does not support clustering nominal variables, should they be re-coded using surrogate keys?

Any help would be really appreciated.

 

Kind Regards

RJ


Accepted Solutions
Solution
‎04-08-2017 05:09 AM
SAS Employee
Posts: 67

Re: Clustering nominal variables in Enterprise miner 14.2

Posted in reply to Ribbonjovi

EM Cluster Node does support nominal variables. The variables need to be identified as input in the data source. Use a Segment Profile node found under the Access tab to explore the relationships. For interval variables the Red respresents the population values, the blue represents the segment. For nominal the inner pie represents the population and the outer represents the segment.

 

2017-04-07_13-32-13.jpg

 

 

HP Cluster node currently only runs on interval variables. You would need to change your nominal variables to indicator variables (0,1) to use in HP Cluster.

 

You can use either node on a single-machine mode. Look in the help to see the different settings/options available in both.  Help can be found by clicking on the blue book with a ? icon and then node reference. Each node's help is organized under the tab name. For example HP Cluster is under the HPDM Nodes section.

 

2017-04-07_13-36-22.jpg

View solution in original post


All Replies
Solution
‎04-08-2017 05:09 AM
SAS Employee
Posts: 67

Re: Clustering nominal variables in Enterprise miner 14.2

Posted in reply to Ribbonjovi

EM Cluster Node does support nominal variables. The variables need to be identified as input in the data source. Use a Segment Profile node found under the Access tab to explore the relationships. For interval variables the Red respresents the population values, the blue represents the segment. For nominal the inner pie represents the population and the outer represents the segment.

 

2017-04-07_13-32-13.jpg

 

 

HP Cluster node currently only runs on interval variables. You would need to change your nominal variables to indicator variables (0,1) to use in HP Cluster.

 

You can use either node on a single-machine mode. Look in the help to see the different settings/options available in both.  Help can be found by clicking on the blue book with a ? icon and then node reference. Each node's help is organized under the tab name. For example HP Cluster is under the HPDM Nodes section.

 

2017-04-07_13-36-22.jpg

Occasional Contributor
Posts: 12

Re: Clustering nominal variables in Enterprise miner 14.2

Posted in reply to MelodieRush

Thank You Melodie, your response was very helpful.

 

Maybe you can help me out with these ones too!

 

  1. Is there a node in Enterprise miner that helps convert nominal variables (especially for the ones with high cardinality) to indicator variables (0,1)?
  2. In the segment profile below do you know if there is an option of labeling (or customizing) all the segment inputs simultaneously (I thought it's laborious to double click each input and editing their graph properties individually)?Untitled picture.png

Really appreciate your help.

 

Cheers

Rj

☑ This topic is solved.

Need further help from the community? Please ask a new question.

Discussion stats
  • 2 replies
  • 705 views
  • 1 like
  • 2 in conversation