Label

Definition

A label is the class assigned to an example.

Example:

input: email
label: spam

or:

input: [hours studied, attendance]
label: pass

Label = the class an example belongs to.

---

Label vs Prediction

A label is the expected or known answer.

A prediction is the model's answer.

Example:

label: spam
prediction: not spam

The model is wrong in this case.

Label is target output. Prediction is model output.

---

Labels In Training

During training, examples usually come with labels.

Example:

input: [2, 5]
label: class A

The model uses the label to check whether its prediction is correct.

If prediction and label differ, the model may update its parameters.

Labels provide feedback for learning.

---

Label Encoding

Labels are often encoded as numbers.

Example:

not spam = 0
spam = 1

or:

class A = -1
class B = 1

The number is not the class itself.

It is a representation of the class.

Numeric labels are encodings of categories.

---

Binary Labels

For binary classification, there are two possible labels.

Example:

0 or 1

or:

-1 or 1

Perceptrons often use -1 and 1 because the update rule becomes simple.

Binary classification uses two labels.

---

Correct And Incorrect Prediction

The model is correct when prediction matches label.

Example:

label: 1
prediction: 1

The model is incorrect when they differ.

Example:

label: 1
prediction: -1

Learning depends on comparing prediction to label.

---

Why Labels Matter

Labels connect features to desired outputs.

Example:

features -> label
[hours studied, attendance] -> pass

Without labels, the model does not know which class each training example should belong to.

Labels tell the classifier what it is supposed to learn.

---

Final Summary

• A label is the known or target class for an example.
• A prediction is the model's output.
• Labels provide feedback during training.
• Labels can be encoded numerically.
• Binary classification has two possible labels.
• Perceptrons often use labels -1 and 1.
• Learning compares predictions with labels.