Interpreting gini

[Scott86]

Hi Everyone,

I have build a decision tree for each month identifying customers who go bad. From April to May the gini coefficient jumps from 30% to around 50%. I think it is largely drive by one node in the decision tree which is high risk where the bads in that node go from 16% to around 50% from April to May. Would this line of thinking be correct or is the gini coefficient for this node around 0 because 50% of the obs are bad and the other 50 are good? I am looking at the equation for gini coefficient in the link below and am not wrapping my head around it.

https://documentation.sas.com/?docsetId=emref&docsetTarget=p1qzwz7onopjqcn11uc04i18urg7.htm&docsetVe...

Any help would be appreciated.

Thanks

[fierceanalytics]

Scott86,

When decision tree is used, one can do predictive modeling (using the past to predict future) or one can describe to get insights.... In the former, one builds a tree using data from a snapshot time (which in your case is a month), then to apply the tree to score on future months, unlike what you are doing. What you are doing essentially is describing monthly variations using a decision tree. In this light, if your job is to report on monthly variations, I would say the explanations details you have provided are easier to understand and more useful than comparing GINI. In other words, the route does not need to go through GINI. GINI is typically if you are scoring a model into future periods and you are looking for a summary way to tell you if the model is off track and needs a rebuild.... Hope this helps. Jia