Talk:One-class classification

Wiki Education Foundation-supported course assignment
This article is or was the subject of a Wiki Education Foundation-supported course assignment. Further details are available on the course page. Student editor(s): Mcastle626. Peer reviewers: Mcastle626.

Above undated message substituted from Template:Dashboard.wikiedu.org assignment by PrimeBOT (talk) 01:51, 18 January 2022 (UTC)

Contradiction?
The opening sentence, "One-class classification, also known as unary classification, tries to distinguish one class of objects from all other possible objects" seems self-contradictory and doesn't make sense from a set theory standpoint. Is one-class classifier a misnomer? Distinguishing "one class of objects" from "all other possible objects" defines two sets. Think of a Venn diagram with just one circle. The circle divides the superset plane (all objects) in two: inside the circle and outside the circle. The objective in classification is to determine where the boundaries between sets - the circles - lie on the plane and so determine to which set points on the plane should be assigned.

Perhaps a better name would be one boundary classifier? --p.r.newman (talk) 11:39, 30 April 2013 (UTC)

Yes, IMHO
Agreed. A unary classifier attempts to classify an instance as belonging to the class or not belonging to the class. There is no assumption of mutual-exclusivity with respect to some class of "all other objects". In the set-theoretic terms, regardless of whether an object belongs to a given subset of the universe, it will always belong to the universe.

So I'd say this is a fundamental mis-representation which should be addressed, by careful rephrasing.

--Justin Washtell — Preceding unsigned comment added by 2.25.231.200 (talk) 18:33, 17 September 2013 (UTC)

Ok, have fixed this now, by rephrasing as follows "One-class classification, also known as unary classification, tries to identify objects of a specific class amongst all objects", and later emphasizing the word "distinguish" in contrast to "identify".

--Justin Washtell — Preceding unsigned comment added by 2.25.231.200 (talk) 18:45, 17 September 2013 (UTC)

Proposed merge with PU learning
It seems that these two problems are very strongly related: PU learning is semisupervised one-class learning. Q VVERTYVS (hm?) 10:10, 3 March 2014 (UTC)

The inputs of one-class classification and PU learning should be different
One-class classification: just need examples from 1 class PU learning needs the example from 1 class (positive) and also unlabeled data (which have hidden positives and hidden negatives) — Preceding unsigned comment added by 192.122.131.37 (talk) 05:43, 11 February 2015 (UTC)

Another vote for merging. I too suggest you merge these two articles, and point out the subtle differences with text. — Preceding unsigned comment added by 24.113.90.243 (talk) 23:33, 24 March 2015 (UTC)


 * I've merged the two. Q VVERTYVS (hm?) 20:00, 16 April 2015 (UTC)