Usage of WoE or Grouping Variables In Regression Node

[SIlver_Geo]

Hello, I am new in SAS Miner and I am trying to understand the following:

1)Why when I use an interactive group node I have as an output both Grouping and WoE variables?

2)Why the Regression Node uses both group and woe variables in the final model?

I am trying to do a coarse classification and then a logistic regression.

I am attaching a screenshot of the project, Interactive Group node & Regression Node settings and the output of the regression. Am I doing something wrong?

Thank you in advance!

[M_Maldonado · Accepted solution]

Hi Silver Geo,

If I understand the Interactive Grouping and Scorecard nodes correctly,

1) You get both the groupings and the weights of evidence in case you have a preference to fit a logistic regression using either of them. Notice that the fourth train property in the Scorecard node is Analysis Variables and you can specify it as WOE (by defualt) or Group.

Most SAS Enterprise Miner customers would use the WOE.

2) The Scorecard node is set to automatically use either the WOE or Groups. This is not the case with the Regression node. The Regression node gives you the flexibility of additional options like the type of regression and the link function, but you have to also take control of the inputs that you use. A case where you might like to take this approach would be when you are interested in a linear regression, but you would have to use either the WOE or the Group variables as inputs.

For your coarse classification and logistic regression, I would advise using the Interactive Grouping node and the Scorecard node. These two nodes are well suited for logistic regression and you also get a scorecard format with points that are easy to interpret. By default, every 20 points the odds of an event double.

I hope it helps,

Miguel

[SIlver_Geo]

Dear Miguel,

Thank you for your assistance, I made quite an extensive reading and I was able to understand the whole process. I used in the end only WoE variables since in behavioural credit scoring can make more sense and show the risk for every variable.

Best Regards,

George