I have a dataset to predict whether applicants default on a loan.

Within the dataset, I have a DURATION variable that works as a time variable so I could perform survival analysis.

The thing is, I want to know if I could use a classification model and not survival analysis on the data to predict DEFAULT. Thoughts?

...or can survival analysis be used to predict binary: someone defaulted/did not default?


1 Answer 1


A binary (e.g., logistic) regression (perhaps as a precursor to later classification) could work if you have complete data covering the exact same period of time for all cases. Then you would model the probability of default over that specific period of time.

In general, though, if you have time information available it's best to do survival analysis. That way you can predict the probability of default as a function of both time and covariate values. That provides more flexibility in downstream use of your model. For example, with time information you could model the actual expected loss by combining the estimated time to default with the payments up through that time point. A simple defaulted/didn't binary model can't do that.

  • $\begingroup$ I am a bit new to survival analysis. So are you saying that I could end up with a prediction that says "new observation defaults after 'x' amount of months? Do you know if ctree() from library(partykit) does that? $\endgroup$
    – Antonio
    Dec 9, 2022 at 0:08
  • 1
    $\begingroup$ @Antonio in general, the prediction would be the probability of "survival" (no default) as a function of time, given a set of covariate values. That's called the survival function. For a tree model like you get from ctree(), each terminal node would have a separate survival function. See Figure 6 of the ctree vignette. My sense is that boosted trees work better than simple trees for prediction in survival models, but it's not so easy to infer specific covariate effects with them. $\endgroup$
    – EdM
    Dec 9, 2022 at 2:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.