Threshold

Definition

A threshold is a cutoff value.

It turns a score into a decision.

Example:

if score >= 0:
  output = 1
else:
  output = -1

Here the threshold is:

0

Threshold = cutoff for deciding between outputs.

---

Simple Example

Suppose:

threshold = 10

Decision rule:

if score >= 10:
  pass
else:
  fail

Scores:

12 -> pass
9  -> fail
10 -> pass

Threshold separates one decision region from another.

---

Threshold In A Perceptron

A perceptron computes a score:

score = dot(w, x) + b

Then it applies a threshold rule:

if score >= 0:
  class 1
else:
  class -1

This converts a number into a class.

Threshold turns a linear score into a classification output.

---

Threshold vs Bias

Threshold and bias are closely related.

Rule:

dot(w, x) >= threshold

can be rewritten as:

dot(w, x) - threshold >= 0

So:

b = -threshold

This is why many formulas use threshold 0 after including bias.

Bias can move the threshold into the score formula.

---

Threshold Creates A Boundary

The threshold defines where the decision changes.

Example:

score = 0

is the boundary between:

score >= 0
score < 0

For linear models, this boundary becomes a line, plane, or higher-dimensional flat surface.

Threshold creates the decision boundary.

---

Too High Or Too Low

Changing the threshold changes decisions.

Example:

threshold = 0

may classify many inputs as positive.

threshold = 100

may classify fewer inputs as positive.

Threshold controls how much score is required to choose a class.

Threshold changes how strict the decision rule is.

---

Why Threshold Matters

Without a threshold, a weighted sum is just a number.

With a threshold, the number becomes a decision.

Example:

weighted sum -> score
threshold -> class

Threshold is the bridge from score to class.

---

Final Summary

• A threshold is a cutoff value.
• It turns a score into a decision.
• Perceptrons use threshold rules.
• Bias can absorb the threshold into the score.
• The threshold creates a decision boundary.
• Changing the threshold changes classification behavior.