community
cancel
Showing results for 
Search instead for 
Did you mean: 

Alteryx designer Discussions

Find answers, ask questions, and share expertise about Alteryx Designer.

Using the Decision Tree node

Atom

Hi all,

 

I'm using the "Decision Tree" node and it's not enough clear for me how to fit the "Complexity Parameter" in order to get the most reasonable number of branches in my model.

 

Moreover, it's still a doubt how to use the output report, once the tree is defined, to insert in it another data set to test the model. I was guessing it could be directly, but it seems we have to know how the tree was built-up, and then to take this criteria into a "Formula" node to split the test set and then to evaluate each register.

 

Probably the most direct way to contribute to my questions is to share an example you handled before.

 

Thanks a million in advance!

Meteoroid
Sounds like you should have attended the training at inspire on predictive modelling.
First step is to cleanse and prep and understand your data using the investigative tools. Field summary is really useful to find missing values etc
Then you can split the data so that you have a training and validation data set.
Then run the varios models and compare the outputs to find the best model, you can use the score to see how it performs ..
There are also some great examples on the help menu for predictive models... hth
H
Atom

Thank you!

 

I have already enough knowledge on predictive models and I know all the details to cleanse, to split by entropy criteria, Gini iindex or whatever, and even to understand how the flow will work and the results; that was not the discussion.

 

The question was focused on something further, which is what's the theory behind the value of the complex parameter to understand with that something about the tree structure (prunning associated, number of levels, how to control sensitivity, specificity...); and the way to replace the standard nodes by other constructions to explain the model better to an audience without technical knowledge.

 

Thanks to your contributions in other posts based on your examples I could get it, so thank you very much to all!!

 

 

Labels