Simple Perceptron (From Scratch in JavaScript)

A minimal implementation of a single-layer Perceptron written from scratch in pure JavaScript.

This implementation demonstrates how a basic linear binary classifier works using randomly initialized weights and bias, trained via the Perceptron learning rule.

---

📌 Overview

This project implements a:

• Single neuron (Perceptron)
• Binary classification model (2 classes: 0 and 1)
• Step activation function
• Supervised learning using labeled training samples

The model initializes weights and bias randomly and iteratively updates them to minimize classification errors.

---

🧠 How It Works

1️⃣ Initialization

• Random weights
• Random bias
• Defined number of input features

2️⃣ Forward Pass

The output is computed as: output = step( w · x + b )

Where:

• w = weights
• x = input features
• b = bias
• step() = activation function

3️⃣ Activation Function

Binary step function: if (sum > 0) return 1; else return 0;

This makes the model suitable only for binary classification problems.

---

🔁 Training Algorithm

The model is trained using the Perceptron Learning Rule: w = w + learning_rate * error * input b = b + learning_rate * error Where: error = target - output

Training continues for a fixed number of iterations or until no weight updates occur.

---

⚠️ Limitations

• Works only for linearly separable data
• Supports binary classification only
• Not suitable for regression problems
• Uses non-differentiable step activation

---

🚀 Usage Example

const p = new Perceptron(2);

p.train(trainingInputs, targets, 0.01, 1000);

const result = p.calculateOutput([x1, x2]);

---

📚 Educational Purpose

This implementation is designed for learning and understanding:

Linear classifiers

Decision boundaries

Weight updates

Fundamental neural network concepts