Search code examples
spssspss-modeler

Splitting continous variables in more then 10 groups in CHAID

I'm using SPSS Modeler 16.0 and i'm trying to make a decision tree using the CHAID algorithm, but the default maximum of nodes in one branch for continous variables is 10 and i would like to have more.

How can change the default of 10?

enter image description here

Here there are only 5 nodes, i would like to have more then 10.

Solution

  • The binning of continuous predictors is based on an algorithm that is described in detail on page 72 of the "Algorithms Guide" 1 and it is correct that the algorithm will attempt to create up to 10 bins, but can end up with fewer depending on the actual distribution of the data and the predictive power of the field.

    If you think that a different binning should be used in order to improve your model, you can do so before passing the data into the model trainer by using the "Binning" node from the Field Ops palette. This will allow you to bin the fields in pretty much any way that you would like. you can then use the pre-binned version of your field in the model, but remember to either filter out the original field or set its direction to 'None'.