My dataset has many biomarkers and the boxplots of these variables show the presence of many outliers. However, these 'outliers' are real data and not misread observations. I want to use elastic net logistic regression to see the association of these biomarkers with a binary outcome. Because my sample size is small and some biomarkers are correlated, I need some regularization. However, I'm not sure if the presence of outliers is a problem for elastic net logistic regressions (I assume it's going to affect the fit adversely because the estimation is still likelihood based). I could only find an R package called enetLTS that does a robust version of the elastic net regression.

I'd like to know if it's better to use this technique or there are other methods that I didn't find. Is there a package that can help identify the outliers and influential observations appropriately so that I can remove those and rerun the regression (if that is suggested at all)? I have found a link that shows sometimes fitting a GLM with the selected variables fails to identify the outliers.


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.