I have 3 sets of data.

  1. A positively labeled dataset.
  2. An unlabeled dataset that has for sure positive (around 75%) and negative data.
  3. An unlabeled dataset that has for sure positive data and maybe negative data.

I'm interested in finding negative observations in this last dataset.

There will be no future new (test) data.

Im trying out some semisupervised logistic regression techniques using shift parameters (http://arxiv.org/abs/1305.4987).

  1. Since I wont have future data should I include all my samples in building my classifier and classify my training data to determine the labels of dataset 2? Or should I train my classifier on only dataset 1 and 2 and then determine the labels of dataset 3?

  2. If I include all data would I still have a problem with overfitting? Do I still need L1 regularisation or something?

  3. Is my choise of training algorithm appropiate for this problem?
  4. Im also strugeling what a good performance measure would be. Missclassification error is probably not ideal.

best regards

  • $\begingroup$ Do you mean that in the 1st data set there are only positive labels and no negatives at all? $\endgroup$ – Alexey Grigorev Aug 7 '14 at 17:17
  • $\begingroup$ Yes, all these samples are almost for sure labeled correctly as positive. $\endgroup$ – statastic Aug 7 '14 at 18:30
  • $\begingroup$ I just came across this wiki page - en.wikipedia.org/wiki/One-class_classification. Maybe it is what you're looking for? (also check the references) $\endgroup$ – Alexey Grigorev Aug 21 '14 at 14:24

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.