Making a Prediction

From Input to Prediction

We have already followed the input through the network. The complete flow looks like this:

Input
  ↓
Hidden Layers
  ↓
Output Layer
  ↓
Output Value
  ↓
Prediction

The important point is that the output value is not always directly the final class or answer. Its meaning depends on the type of problem.

Example 1 — Binary Classification

Suppose we build a neural network that predicts whether a student will pass or fail.

Input:
Study Hours
Attendance

Output:
Pass or Fail

Suppose the output layer uses Sigmoid and produces:

Output = 0.889

For a binary classification problem, this can represent the model's predicted probability of the positive class, assuming the model was designed that way.

0.889 = 88.9%

If we use 0.5 as the decision threshold:

0.889 >= 0.5

Prediction = Pass

Another Binary Example

Suppose another student produces:

Output = 0.23

Using the same 0.5 threshold:

0.23 < 0.5

Prediction = Fail

So the model's output can be converted into a class using a decision rule.

Output >= 0.5 → Positive Class
Output <  0.5 → Negative Class

The threshold of 0.5 is common, but it is not a universal rule. The appropriate threshold can depend on the application and the costs of different errors.

Example 2 — Multi-Class Classification

Now suppose a neural network needs to identify an animal in an image.

There are three possible classes:

Cat
Dog
Bird

The output layer might produce:

Cat  = 0.80
Dog  = 0.15
Bird = 0.05

The largest output is:

Cat = 0.80

Therefore, the predicted class is:

Prediction = Cat

Why Do We Get Multiple Values?

For a multi-class problem, the output layer commonly uses Softmax when the classes are mutually exclusive.

Softmax converts the output scores into values that can be interpreted as probabilities across the classes.

Cat  = 0.80
Dog  = 0.15
Bird = 0.05

Total = 1.00

The class with the highest probability is commonly selected as the predicted class.

Example 3 — Regression

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

For example, suppose we want to predict a house price.

Input:
House Size
Bedrooms
Location

Output:
House Price

Suppose the network produces:

Output = 250000

The prediction is:

Predicted Price = $250,000

There is no "choose the largest class" step here. The output itself is the predicted numerical value.

The Output Depends on the Problem

So you cannot look at a neural network output and automatically assume it always means the same thing.

Complete Example — Student Prediction

Let's follow one student through the complete network.

The input is:

Study Hours = 5
Attendance = 90

After the hidden layers, suppose we get:

Hidden Output:

[3.0, 2.0, 0.9]

The output layer calculates:

z = (3.0 × 0.4)
  + (2.0 × 0.3)
  + (0.9 × 0.2)
  + 0.1

z = 2.08

Apply Sigmoid:

Sigmoid(2.08) ≈ 0.889

Now interpret the result:

0.889 = 88.9%

88.9% >= 50%

Prediction = Pass

Prediction vs Probability

These two ideas are related but they are not exactly the same.

Model Output:
0.889

Probability:
88.9%

Decision:
Pass

The neural network produces the output value. A separate decision rule can then turn that value into a class.

This distinction becomes important when you later learn about classification thresholds and model evaluation.

Making a Prediction With Python

We can convert the model output into a simple binary prediction:

prediction_probability = 0.889

threshold = 0.5

if prediction_probability >= threshold:
    prediction = "Pass"
else:
    prediction = "Fail"

print("Probability:", prediction_probability)
print("Prediction:", prediction)

Output:

Probability: 0.889
Prediction: Pass

Multi-Class Prediction With Python

Suppose the model gives us three class probabilities:

probabilities = {
    "Cat": 0.80,
    "Dog": 0.15,
    "Bird": 0.05
}

We can select the class with the highest probability:

prediction = max(
    probabilities,
    key=probabilities.get
)

print(prediction)

Output:

Cat

What Happens After a Prediction?

During inference, we can simply use the prediction. During training, however, we compare the prediction with the actual answer.

Input
  ↓
Forward Propagation
  ↓
Prediction
  ↓
Compare With Actual Answer
  ↓
Calculate Loss
  ↓
Backpropagation
  ↓
Update Weights

The loss and backpropagation steps are part of the training process and will be covered later.

The Big Picture

Input
  ↓
Hidden Layers
  ↓
Output Layer
  ↓
Raw Output
  ↓
Activation Function
  ↓
Model Output
  ↓
Decision Rule
  ↓
Prediction

The exact final step depends on the machine learning problem.