Details about the data source: credit card churn data
It has only 5% churn data. Because of this rare event, tried changing the decision weights at the data source.
Here is a screenshot of the tree:
I have taken care of the quasi-separation/separation issues by excluding them from the metadata node which is then connected to the decision tree node.
1. However, the results does not show a proper split between churn and non-churn. How to overcome this problem? Kindly help. Thanks
2. Also, wanted to check if choose to change the decision weights at the data source, should the assessment measure in the decision tree properties(panel) only be 'DECISION'? or can I use other assessment measures? What does the LIFT measure mean? I noticed that using the lift measure helped to improve the distribution of the churn and non-churn in my best nodes. But am not sure what the LIFT measure means.
3. Under the split search option for decision trees, should 'use decision' be set to Yes since decision weights were selected at the data source?
Kindly help with the above doubts! really appreciate it!
![Ann_90_1-1585879526478.png Ann_90_1-1585879526478.png](https://communities.sas.com/t5/image/serverpage/image-id/37726iD383FB5BD18E0AE1/image-size/medium?v=v2&px=400)
![Ann_90_2-1585879552751.png Ann_90_2-1585879552751.png](https://communities.sas.com/t5/image/serverpage/image-id/37727iB5B90D0A5E6EBF67/image-dimensions/640x88?v=v2)