Which MATLAB function is the best for building a decision tree with the CART algorithm?

Hello there, I want to build a tree using the CART Algorithm and so far I found two different (?) functions in the Matlab statistics toolbox for doing this: ClassificationTree.fit and classregtree, so I am wondering which of them is better or whether they are both based on the same principles, but with different application fields?

Answer 1

[owr]

I believe they are using the same algorithms. "classregtree" has been around for quite some time, "ClassificationTree.fit" is syntax based on a newer object based framework. Note I havent researched this rigorously, just a hunch.

If I were writing new code, I would go with the object based syntax as that will likely get more bells and whistles down the line.

[Tom Lane]

Hunch is correct.

Answer 2

[Muhammad Aasem]

use classregtree because it will be supported in the future. anyway. both will give you same result (treefit is now calling classregtree)

try this

load fisheriris;
t1 = classregtree(meas,species);
t2  = treefit(meas,species);
view(t1);
view(t2);

[Ines]

but how can I assess the goodness of the qualification? I would like to take out 20% for validation and use the residual 80% for the training dataset..and then repeat this procedure several times...i think if I use crossval I cannot use a tree from classregtree (since crossval seems to ask for an object/handle (whatever that might be..)