I am trying to predict the termination events for a company. I also want to rank the most important variables that lead to termination. I ran a decision tree model and have the following questions:

  1. Is the variable importance (respective ranking of variables) of decision tree reliable, when the overall accuracy, sensitivity, specificity and KAPPA of the model is low?

  2. Decision trees tend to favor variables with high variance. Does this imply that numeric variables will mostly land up higher in the list of variable importance?

  • $\begingroup$ First, there are many different types of "decision trees", different algorithm will have different ways to select variable to split, select threshold for a given variable, and have different ways to define variable importance. In R, RPART, CTREE, PARTY are different. Are you talking about Breman's CART? (RPART in R)? $\endgroup$ – Haitao Du Jul 1 '16 at 18:08
  • $\begingroup$ @hxd1011 yes it is rpart in R $\endgroup$ – Batool Jul 1 '16 at 18:25

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.