Reference or deeper details for the CRT TREE algorithm

[Tim Graettinger]

Hi,

I'm trying to understand the details of the SPSS CRT Decision Tree algorithm, particularly the splitting criteria.  I managed to find a document on the IBM site entitled, IBM SPSS Decision Trees 25 that talks about the CRT Criteria: impurity measure, Gini, Twoing, and Ordered Twoing.  What are really missing for me are some examples with the calculations shown, or at least the formulas that are being applied.  I've tried to do some of my own examples by hand, but haven't been able to reproduce the impurity/improvement measure shown in the results.

If anyone has a pointer to the original paper(s) on which the SPSS implementation is based, that would be very helpful as well.  Thanks!
-Tim

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Jon Peck]

Have you looked at the TREE documentation in the Algorithms manual

ftp://public.dhe.ibm.com/software/analytics/spss/documentation/statistics/26.0/en/client/Manuals/IBM_SPSS_Statistics_Algorithms.pdf

On Tue, Apr 23, 2019 at 7:18 AM Tim Graettinger <[hidden email]> wrote:

Hi,

I'm trying to understand the details of the SPSS CRT Decision Tree algorithm, particularly the splitting criteria.  I managed to find a document on the IBM site entitled, IBM SPSS Decision Trees 25 that talks about the CRT Criteria: impurity measure, Gini, Twoing, and Ordered Twoing.  What are really missing for me are some examples with the calculations shown, or at least the formulas that are being applied.  I've tried to do some of my own examples by hand, but haven't been able to reproduce the impurity/improvement measure shown in the results.

If anyone has a pointer to the original paper(s) on which the SPSS implementation is based, that would be very helpful as well.  Thanks!
-Tim

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

--
===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

[Tim Graettinger]

Thanks, Jon!  Just what I was looking for!!
-Tim

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD