Output Layer

Where Is the Output Layer?

The output layer is always at the end of the neural network.

Information flows from the input layer, through the hidden layers, and finally reaches the output layer.

What Does the Output Layer Do?

The output layer converts the information learned by the network into a result that we can use.

The exact output depends on the machine learning problem.

Example 1 — Pass or Fail

Suppose we want to predict whether a student will pass an exam.

The input could contain:

The information moves through the network:

The output layer produces the final result for the student.

Binary Classification

Binary classification means there are two possible classes.

Examples include:

A common output is a value representing the probability of one of the two classes.

Output = 0.92

This could mean that the model estimates a 92% probability for the positive class, depending on how the model and labels were defined.

Multi-Class Classification

Sometimes a model needs to choose between more than two classes.

For example, suppose we want to classify an animal:

The output layer can contain multiple outputs, with each output corresponding to a class.

The largest probability is for Dog, so the model's predicted class would be Dog.

Regression Output

Not every neural network predicts a class. Some neural networks predict a numerical value.

For example, we might predict the price of a house.

In this case, the output layer produces a numerical value instead of "Cat", "Dog", "Pass", or "Fail".

Prediction = 350000

The Output Layer Depends on the Problem

There is no single output-layer structure that works for every machine learning problem.

The problem determines what the output should represent.

How Many Neurons Does the Output Layer Have?

The number of output neurons depends on the task.

For a simple binary classification problem, a common design uses one output neuron.

Output Layer
    ↓
    Neuron

For a three-class classification problem, a common design uses three output values.

Output Layer
    ↓
    Neuron 1 → Cat
    Neuron 2 → Dog
    Neuron 3 → Horse

For a regression problem that predicts one number, a common design uses one output value.

Output Layer
    ↓
    Neuron → House Price

Output Layer vs Hidden Layer

Complete Example

Let's put everything together using the student example.

For example, the network might produce:

Prediction = 0.92

The application can then interpret that output according to the problem and the chosen prediction rule.

The Output Is Not Always a Word

A common beginner mistake is thinking that the output layer directly produces words such as "Cat" or "Dog".

Internally, the neural network usually produces numerical values. Those values are then interpreted according to the task.

For example:

0.85
    ↓
Dog