Build a CRT model.
CRT trees are sometimes better built “intuitively”. While there is a KPI that lets us know where to prune, there is a subjective interpretation that is better not to leave to algorithm.
This node generates a tree and give a web application in a browser in which you can choose where to cut.
Make sure to specify the two models: Initial and Final/Pruned models. Those to model files will be saved autonatically. For more information about the CRT model, see the sections 5.12.2. and 5.12.3.
AUC automatically adapts to the new selection. Once you see the AUC is stable between learn and test, you probably have a good model.
If the target is continous, instead of the AUX, the model will compute the RMSE and R2
If the target is multinomial, the plots will be a classification table.
You can zoom on the tree to see more details, and on-MouseOver, more details will display about the node.