1
$\begingroup$

I was reading about One Class Classification but had two doubts -

1) How does One Class Classification work because the training data is of only a particular class so in that case, during testing, the model would always predict that any test data belongs to that class only.

2) And since it does work then is it not perfect for Spam Detection because we can have data for spam emails but how could we find any amount of training data on Non Spam to possibly cover all cases for Non Spam E-mails. As any general email is a non spam one and only certain mails are spam.

$\endgroup$
  • $\begingroup$ Do you mean "One Class Classification" in the context of Outlier Detection like OneClassSVM? $\endgroup$ – TwinPenguins Jun 12 '18 at 13:29
  • $\begingroup$ No what I meant was more in lines of building a repository of positive class documents in such a case positive documents are readily available and then one can use it to find same class of documents from any other sources without manual labelling of negative documents from each source. $\endgroup$ – Ojasvin Sood Jun 12 '18 at 13:57
  • $\begingroup$ This is the example I found in a research paper titled “ Partially Supervised classification of Text Documents “ $\endgroup$ – Ojasvin Sood Jun 12 '18 at 13:59
  • $\begingroup$ Yes, I see but this to me also sounds like outlier detection or novelty detection in an unsupervised fashion. People are using only one class (positive like your example) in autoencoders and based on reconstruction error they will label new incoming inputs as either positive or negative. See this example: medium.com/@curiousily/…. Unless I do not understand yet your Q. $\endgroup$ – TwinPenguins Jun 12 '18 at 14:12
  • $\begingroup$ It sounds like an a anomaly detection problem. I guess technically not a spam itself is a class, so it is still somewhat binary. The different is to detect features that makes one case much different from the other $\endgroup$ – The Lyrist Jun 12 '18 at 14:14

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.