Interactive Gouping Help

[MarcosEilert]

Hi, I need help on the node "Interactive Grouping" from Sas Enterprise Miner.

To be more specific, I need to understand what does the option "Smallest Branch with Residual" from "Missing Values" Parameter.

What is the theoretical background of this option? How works the algorithm?

Someone have any book suggestions or other content that can help me answer these questions?

Thank you.

[M_Maldonado]

Interactive Grouping has the option to use two approaches to come up with the bins or groups for each of your input variables. One way is to use a decision tree using proc arbor, the other is an operation research approach using proc optbin.

When you use a decision tree through the options Optimal Criterion, Quantile, or Monotonic Event Rate,  you have all the tree based grouping options that you would expect.

If you are familiar with proc arbor, the small residual option assign missing values to the leaf so that the Sum of Square Error is the lowest. It might have a different name in other decision tree algorithms.

You can find more information on proc arbor and its missing value handling in this document. Missing value options are discussed in page 17. http://support.sas.com/documentation/onlinedoc/miner/em43/allproc.pdf

I hope it helps,

Miguel