0

Is it possible to do one-class classification (wiki) with Microsoft ML.NET? I.e. in the training data set there is only one (the positive) class and I would like to tell if some test data belongs to this class or not. Can this be accomplished wiht ML.NET?

Pobably I can try with binary classification (although the prediction will always be classified as that class in the training data set) and inspect the score of the prediction, but this doesn't seem like a robust solution.

Thank you.

ripe_bananas
  • 1,060
  • 9
  • 13
  • We can use the tool to help we do it. See https://marketplace.visualstudio.com/items?itemName=MLNET.07 – lindexi Oct 29 '19 at 12:01
  • Thanks for the suggestion, but I am not looking for a tutorial how to train a model (there's plenty out there). I am asking what trainers/algorithms in ML.NET can be used to do OCC. And if none - if there is a workaround. – ripe_bananas Oct 29 '19 at 13:54
  • just want to say that this is a very nice question and I am now stuck with the same problem, it seems that not so many studies have covered well this problem.. did you found a workaround at last! – Warios Jun 17 '21 at 22:07
  • No, not really. – ripe_bananas Jul 02 '21 at 11:46

1 Answers1

1

Binary classification will not work with only one class, since the model will devolve into constant prediction.

We used to have one-class SVM in the internal precursor to ML.NET, but I don't see any evidence that it made it into the library. In this case, no, I cannot think of any workaround using existing algorithms.

Zruty
  • 8,377
  • 1
  • 25
  • 31