Several books that I have read do not distinguish the several models that exist for anomaly and outlier detection.

After I read about these models, I have chosen to detect anomalous events on unsupervised and supervised data the following algorithms:

Unsupervised data:

  1. K-means
  3. One class support vector machine

Supervised data:

  1. Support vector machine

Let's imagine that these models were really good at detecting anomalous points. Now I want to detect outliers and novel events. Can I use the same algorithms? If so, in what sense it differs using anomaly, outlier, and novelty detection models?

In outlier and novelty detection I need to remove anomalous points before applying a model?


1 Answer 1


As far as I know they all refer to the same problem. Put differently: anomalies are treated as novelties, that is, something your model of the pdf assigns a very low probability.

Following this idea, if your model is an unsupervised model which fits some model for the pdf of your data, then yes, you need to set the outliers apart, and then test on them. If, on the other side, you use a supervised model, then you need to consider all your data.

  • $\begingroup$ Anomalies can be treated as novelties? Are they the same thing? What do you mean by saying that the model fits some probability density function of the data? Are you referring to data that is normally distributed? $\endgroup$
    – xeon123
    Mar 3, 2020 at 9:19
  • $\begingroup$ my point is: they are treated in the same way. There are events that, according to some distribution of probability, are very unlikely to happen. So you end up using the same tools to deal with them $\endgroup$
    – jpmuc
    Mar 3, 2020 at 17:41
  • $\begingroup$ @jpmuc May I draw your attention to this post kindly? $\endgroup$
    – Mario
    Apr 21, 2022 at 6:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.