# What is the difference between a multi-label and a multi-class classification?

What is the difference between multi-label classification and multiclass classfication. Speficially, what is the difference between a label and a class? Please provide a clear example.

"Multiclass classification should not be confused with multi-label classification, where multiple labels are to be predicted for each instance." -wikipedia ... was not very helpful.

Based on the sentence you quoted, each item belongs to one class but can have several labels.

Imagine you have animals like a fox, a chicken and a common European viper. A multi-class classification problem would be assigning them to a family:

Fox            Canidae
Chicken        Phasianidae
Viper          Viperidae


In phylogeny, any species only has one family (that's by design) so that an animal cannot belong to more than one family.

A multi-label classification problem would be assigning them random characteristics:

Fox            Warm-blooded, furred
Chicken        Warm-blooded, feathered
Viper          Cold-blooded


Each animal can have several labels and the labels do not form a set of mutually exclusive categories.

• Oh, I hadn't thought of modeling a problem like that. The quote I provided was the only instance (sadly) describing the two together that I could find. In this instance, its seems that a class is entered into the multi-label classification network, whose inputs should be able to handle the feature-set for all types of desired animals. Understanding the difference between a label and a class was the most help. If there are other types of multi-label models other than the one described in your answer (which sounded like it was based on my quote), please elaborate. Otherwise, thank you. – poorly_built_human Jul 14 '14 at 2:03
• @poorly_built_human My example was not very good, I tried to use animals in both cases but things like cold-blooded vs. warm-blooded or furred vs. feathered can be treated as separate multi-class classification problems. The canonical multi-label classification problem would be identifying topics in a text. – Gala Jul 14 '14 at 17:09
• Thank you again. I'm seeing classes as generally being mutually exclusive (from an organizational POV), while labels are not mutually exclusive. Likewise, a class is a larger group where an item is either "in it" or not, and a label is an attribute of a class for which many can exist. You might try and classify text into a particular "class", such as History, Science and Literature, and then give labels to these classes such as "Science: Biology" and "Science: Geology". Or even more specific labels (or combination thereof) such as "Science: Geology: Europe" and "Science: Geology: Africa". – poorly_built_human Jul 15 '14 at 7:41