Logistic regression fits a model that predicts a binary variable whilst performing a logit transformation of the linear combination (LC) of predictors: 1/1 + exp(-LC).

I have a working machine learning algorithm which does the same but fits a non linear combination of predictors by maximising the log likelihood. I would like to apply this algorithm to presence only data as suggested here:

Presence-only data and the em algorithm

Could I just use formula (4) (its log likelihood version)? How do I use the resulting fitted model to perform predictions if I do not know the value of z but just the predictors/co-variates? I also do not understand the difference between observed and full likelihood? Any clarification would be very much appreciated.


It turns out that I can use formula 4 but have to use the standard logit link function for predictions.


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.