1

I’m confused, in wikipedia, ordinal regression is also referred as ordinal classification. Which makes sense since ordinal variables are in the end just categorical. On the other hand, ordinal regression is often referred in machine learning as a ranking problem, as we know it from Information Retrieval.

However, if you think about it, ordinal regression is actually not "ranking" (since many values can have the same "ranking") but more a rating, which again is a classical multiclass classification problem?

Justin Lange
  • 354
  • 2
  • 6

1 Answers1

0

See this thread:

Why isn't Logistic Regression called Logistic Classification?

In sum, logistic regression (either binary or ordinal) is not a classification algorithm by itself. You have to specify an arbitrary probability threshold for class membership.

Bálint L. Tóth
  • 413
  • 3
  • 10
  • Check! But your answer still comes back to the same question: when using ordinal regression for classification, why is it often considered a ranking problem? – Justin Lange Jun 12 '19 at 11:21
  • This looks like a clash in definitions: in statistics, "ranking" usually refers to relative ranking among observed values, whereas here it's referring to comparing values against pre-determined thresholds. Rank as in exam grades, not as in tournament placing. – Accidental Statistician Oct 01 '20 at 11:10