0
$\begingroup$

I'm creating credit scoring model and stuck with WoE calculation. I know the formula and I know how to compute WoE for train sample. Should I use train sample WoE for test sample or I should compute new WoE for test sample?

$\endgroup$
  • $\begingroup$ I don't understand the question. What is the issue? $\endgroup$ – Umka Mar 1 '19 at 16:13
  • $\begingroup$ The question is whether to recalculate new WoE for test set or not $\endgroup$ – Andrew Mar 1 '19 at 16:23
  • $\begingroup$ It seems that you want to check the predictive validity of your WoE calculation (Am I right?). In that case you will need to compute WoE for the test sample and also predict the WoE for this sample. Finally you will compare computed (observed) and predicted WoE on the test sample. The train sample will only be used to build your model, which will be used to predict the WoE on test data. $\endgroup$ – Umka Mar 1 '19 at 16:29
  • $\begingroup$ I'm asking about methodology. I think recalculating WoE using test set is incorrect because we don't know numbers of goods and bads in sample apriori. So, train set WoE should be used as long as it represents population proportion $\endgroup$ – Andrew Mar 1 '19 at 19:50
  • $\begingroup$ How do you define train and test samples? (We might used different definitions) In the cross-validation domain, train and test only refer to a partitioning of the full sample (say you randomly split your sample into 2 groups) - Therefore I am unclear why you would know goods & bads for only one of the subsamples $\endgroup$ – Umka Mar 1 '19 at 19:58

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.