BookmarkSubscribeRSS Feed
relish
Calcite | Level 5

Hi, I am using sas enterprise miner to conduct predictive modeling, while I still can not figure out how to identify latent factors in the software, I have browsed the journals and books while still can not find the solution  and can not have the clear understanding about this concept, is there any tutorial or references or how to solve it in sas EM?

11 REPLIES 11
PaigeMiller
Diamond | Level 26

@relish wrote:

I still can not figure out how to identify latent factors in the software

"identify latent factors in the software" -- do you mean what mathematical or statistical procedure to run to compute latent factors? Or do you mean something else?

 

If that's what you mean, there are many procedures that will compute latent factors such as Partial Least Squares regression, Principal Components Analysis, Factor Analysis and probably several others.

--
Paige Miller
relish
Calcite | Level 5

Thanks so much for your reply, Yes, I mean , what mathematical or statistical procedure to run to compute latent factors in sas enterprise Miner. since PCA is used to reduce dimension of the variables, which I want to use in the later predictive models builds, So I still can use the PCA here for latent factors analysis?

PaigeMiller
Diamond | Level 26

Yes, you can use Principal Components, but I strongly recommend not using Principal Components and use Partial Least Squares (PLS) instead.

 

PCA find components (or "latent factors" as you call them) but does not use the response variables to do so, and so it can miss variables that are predictive of the response. PLS does not have this drawback.

--
Paige Miller
relish
Calcite | Level 5

Thanks so much for your reply,you give me so much insights. while I am confused about the concept of the latent variables, should we identify them ourselves, or I just choose the partial least squares to finish identifying? And what is the difference between the PLQ and exploratory factor analysis? I am really appreciate  your patience.

PaigeMiller
Diamond | Level 26

If you are a subject matter expert, you might be able to specify latent variables based upon yoiur knowledge of the subject. Normally, this is a data-based exercise, where the data determines what the latent variables are.

 

The reason I use PLS instead of PCA or factor analysis is because PLS finds latent variables that are predictive of the response (or Y) variable. Neither factor analysis nor PCA does this, and so they may find latent variables that are not the best predictors, and thus are not as predictive as PLS.

--
Paige Miller
relish
Calcite | Level 5

Thanks so much for your reply. I can understand the difference between the PLS and PCA, while as you said, from the PLS results, so the variable importance for projection is where I can find the variables that are predictive to the target variable?(above the threshold)? and is there any other thing that I can conclude from the PLS result? thanks again!

PaigeMiller
Diamond | Level 26

@relish wrote:

Thanks so much for your reply. I can understand the difference between the PLS and PCA, while as you said, from the PLS results, so the variable importance for projection is where I can find the variables that are predictive to the target variable?(above the threshold)? and is there any other thing that I can conclude from the PLS result? thanks again!


The PLS X-Loadings show variables that have high importance in predicting. If the loading has a value not close to zero (in other words a big positive or big negative number), then it is important in predicting. "Big" is relative, you judge "big" by looking at all the X-Loadings in each dimension and seeing which ones are big. You use the DETAILS option to see the X-Loadings.

 

Other than that, there are many benefits of PLS, but its not really clear what you are asking.

--
Paige Miller
relish
Calcite | Level 5

Thanks so much for your reply, I can see the different importance regarding distinct variables. While for my purpose, I want to identify the latent gators and assess their relationship, so I do not know how to complete this work in PLS. By the way, there seems no operation with factor analysis in sas enterprise miner?

 

really appreciate!

PaigeMiller
Diamond | Level 26

The latent factors importance comes from the table showing the percent importance of each latent factor, as shown in this example, in table 89.1.1

http://documentation.sas.com/?docsetId=statug&docsetTarget=statug_pls_examples01.htm&docsetVersion=1...

 

You can see that latent factor 1 explains over 89% of the response, and then second latent factor explains more than 7% of the response. The rest of the latent factors explain less than 1% of the response. This indicates to me that you would want to use either 1 or 2 latent factors in the model.

--
Paige Miller
relish
Calcite | Level 5

thanks so much for your reply, you help me a lot ! while I am still wondering is there any methods that I can assess the latent factors relationships? really really appreciate!

PaigeMiller
Diamond | Level 26

@relish wrote:

thanks so much for your reply, you help me a lot ! while I am still wondering is there any methods that I can assess the latent factors relationships? really really appreciate!


Relationships with what???

 

As I already stated, the latent factors found with PLS are predictive of the Y-variable (as predictive as the data will allow). Isn't that what you are looking for? I have already described how you find out the predictive ability of the latent factors.

--
Paige Miller

sas-innovate-2024.png

Join us for SAS Innovate April 16-19 at the Aria in Las Vegas. Bring the team and save big with our group pricing for a limited time only.

Pre-conference courses and tutorials are filling up fast and are always a sellout. Register today to reserve your seat.

 

Register now!

How to choose a machine learning algorithm

Use this tutorial as a handy guide to weigh the pros and cons of these commonly used machine learning algorithms.

Find more tutorials on the SAS Users YouTube channel.

Discussion stats
  • 11 replies
  • 1271 views
  • 0 likes
  • 2 in conversation