BookmarkSubscribeRSS Feed
🔒 This topic is solved and locked. Need further help from the community? Please sign in and ask a new question.
WY
Calcite | Level 5 WY
Calcite | Level 5

Hello, everyone.

As a new user to SAS EM, I am trying to use the Decision Tree in SAS EM 12.1 on a prediction project, in which a bunch of facts (system status) are input to predict the performance of the system (measured in three Key Performance Indicators (KPI) ). So far I have created a Decision Tree model and obtained some good preliminary results.

However, I found I am confused about the "Use Multiple Targets" setting in the "Decision Tree" module.

My question is -- Does the "Multiple Targets" refers to

  • multiple target variables/columns in the data source (e.g., I have 3 KPIs, which are all set to Target so that they will be predicted);
  • multiple levels of one specific target, which can be chosen insides the "Edit variables ..." of the Desision Tree module and set the "Use" of one target to Yes and others to No

My guess is the second since if I set the "Use Multiple Targets" to NO, then the generated decision tree will only split once then stop.

In contrast to if I set the "Use Multiple Targets" to Yes, then the decision tree will split for several layers.

I am not sure if my guess is correct or not, and wish someone experienced can help confirm/correct me.

Thanks,


1 ACCEPTED SOLUTION

Accepted Solutions
WY
Calcite | Level 5 WY
Calcite | Level 5

Hi, Jonathan:

Thank you very much for your reply !

I have read the EM User Guide, and have performed several experiments. For now, my understanding are:

1. The property : "Multiple Targets" is only useful when users run "interactive decision tree" since they can "switch target" at any time. For a auto-fitted decision tree, "Multiple Targets" has no impact at all -- users have to explicity specify only one variable to be used as the target by change its "Use" property to "YES" in the "Edit Variable" contect menu.

2. If users have multiple targets and want to check the input facts' impact on each of them, they may use "Start Group" and "End Group" modules to realize such a goal. The "decision tree" module does not fit to more than one target variable in one run.

I will appreciate it if you can make a comment or correct my understandings.

Thanks.

View solution in original post

3 REPLIES 3
jwexler
SAS Employee

Hi, I was able to pull a section from the EM User Guide.  It's a great resource!  I use it often!

Decision trees are commonly used for interactive segmentation, usually market segmentation, sometimes followed by predictive modeling in the segments. With multiple targets, the user can select a different target in a leaf that has become sufficiently homogeneous with respect to the original target. The user ends up with one set of rules to display or score new data with later, instead of having to piece together rules from separate trees.

In contrast to segmentation, statisticians use trees as a predictive model in itself. SAS decision trees with multiple targets are not suitable for this because the split search only looks at one of the targets.

In order to use multiple targets, you set the Use Multiple Targets property to Yes. When multiple target variables are used, you use the Variables property of the Decision Tree node to specify which target variable is to be used as the decision variable for the root node of the tree. You do this by setting the Use property to Yes. All other target variables must have their Use property set to No. You can switch targets by clicking on a leaf and then clicking the Switch Targets icon that appears on the toolbar of the Interactive Decision Tree window. When multiple targets are used, Assessment Plots and Assessment Tables are not available.

WY
Calcite | Level 5 WY
Calcite | Level 5

Hi, Jonathan:

Thank you very much for your reply !

I have read the EM User Guide, and have performed several experiments. For now, my understanding are:

1. The property : "Multiple Targets" is only useful when users run "interactive decision tree" since they can "switch target" at any time. For a auto-fitted decision tree, "Multiple Targets" has no impact at all -- users have to explicity specify only one variable to be used as the target by change its "Use" property to "YES" in the "Edit Variable" contect menu.

2. If users have multiple targets and want to check the input facts' impact on each of them, they may use "Start Group" and "End Group" modules to realize such a goal. The "decision tree" module does not fit to more than one target variable in one run.

I will appreciate it if you can make a comment or correct my understandings.

Thanks.

jwexler
SAS Employee

Hi, sounds reasonable, you could generate different models using Start/End Group Processing (like By-Groups).  When you want to change targets interactively in your tree, simply use the Interactive Tree to change targets.  Make sure you prune your nodes below the node you want to change, and then right-click, switch-targets.

Good luck!

Jonathan

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
  • 3 replies
  • 6426 views
  • 3 likes
  • 2 in conversation