Difference Between Multiclass and Multilabel Classification

Difference Between Multiclass and Multilabel Classification

Multiclass Classification and Multilabel Classification are two different types of classification problems. The key difference lies in the number of labels each sample can have and the relationship between the labels. Below is a detailed comparison with examples:

---

1. Multiclass Classification

• Definition: Each sample can belong to only one class, and the classes are mutually exclusive.
• Characteristics:
  • Classes are independent, and a sample can have only one label.
  • The output is a probability distribution (achieved through the softmax function), where the sum of probabilities equals 1.
• Common Scenarios:
  • Image classification: An image is either a dog, a cat, or a bird.
  • Sentiment analysis: A text is positive, negative, or neutral.
• Loss Function:
  • Usually uses the Cross-Entropy Loss function.

Example

Sample	Label
Image 1	Dog
Image 2	Cat
Image 3	Bird

---

2. Multilabel Classification

• Definition: Each sample can belong to multiple classes simultaneously, and the labels are not mutually exclusive.
• Characteristics:
  • Classes can overlap, and a sample can have multiple labels.
  • The output consists of multiple probabilities (achieved through the sigmoid function), with each class having an independent probability.
  • Labels may or may not be correlated, but they are not mutually exclusive.
• Common Scenarios:
  • Image multi-label classification: An image can contain both “dog” and “cat”.
  • Text multi-label classification: A document can belong to both “science” and “education”.
• Loss Function:
  • Usually uses the Binary Cross-Entropy Loss function.

Example

Sample	Labels
Image 1	Dog, Cat
Image 2	Cat, Bird
Image 3	Dog, Bird, Cat

---

Comparison Summary

Feature	Multiclass Classification	Multilabel Classification
Labels per sample	Only one class	Can belong to multiple classes
Relationship between labels	Mutually exclusive	Independent, multiple possible
Output activation function	Softmax	Sigmoid
Loss function	Cross-Entropy Loss	Binary Cross-Entropy Loss
Output format	Single probability distribution (sum = 1)	Independent probabilities per class