SAS EM: decision tree

[ycenycute]

In SAS EM decision tree node, what is the difference between running "train the node" under "Interactive" and directly running the decision node in the diagram workspace? The results do not seem match.

Also, how to do pruning for decision tree? I tried changing Assessment measure from decision to misclassification, the results do not seem differ.

[gcjfernandez]

In SAS EM decision tree node, what is the difference between running "train the node" under "Interactive" and directly running the decision node in the diagram workspace? The results do not seem match.

Answer:

The default setting related to missing value handling is different between Decision tree node and the interactive tree. Interactive tree mode is suitable for building custom tree.

Also, how to do pruning for decision tree? I tried changing Assessment measure from decision to misclassification, the results do not seem differ.

Answer:

For binary target, in the absence of decision weights, both misclassification and decision assessment measures uses misclassification rate as the model selection criterion resulting in no differences.

[ycenycute]

What is the difference in terms of handling missing data between Decision Tree node and the interactive tree? What does it mean by "custom" tree?

[gcjfernandez · Accepted solution]

Decision tree node default setting for handling missing value: Use in search

Interactive tree default setting for handling missing value: A specific branch (branch 2 by default)

Custom tree : At each split point you can choose which variable to use in split and at what split point value you want to split.

                      Also you can choose which branch to prune.

[ycenycute]

What does it mean by use in search and what does it mean by branch 2?

[gcjfernandez]

Use in search:

In decision tree node with binary split, missing value record is assigned to both left and right side of the branch and the log worth . is estimated. Then the missing value record is assigned to the side that gives largest log worth.

branch 2:

In interactive DT window, by default the missing value record is assigned to the right side only.