1
$\begingroup$

Does there exist packages in R or SAS that are designed specially for logistic regression, tree model or other binary classifiers for rare events?

What is the simplest way to deal with rare events if no package is available?

Adding weights? Over-sampling?

$\endgroup$
1
  • $\begingroup$ Could you define "rare events"? It could have different meanings, e.g. atypical observations or extreme values etc., that need different methods. $\endgroup$ – Tim Apr 28 '15 at 8:36
1
$\begingroup$

If you mean rare events by the outcome variable having binary response where other has a very small frequency then it might be that frequency itself is not so important but number of cases. It might also be that there is not enought variation between response and levels of predictor categorical variables or in range of variation of continous predictor variables. If you do have this variation, it means that it is not possible to infer marginal effect of some variable to the outcome.

Firth penalized likelihood estimator can also be used, for example when having small amount of cases and possible some zero cells in multidimensional space when having categorical predictors.

http://cran.r-project.org/web/packages/logistf/logistf.pdf

$\endgroup$

Your Answer

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

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