Handwriting Digit Detection - Binary Classifier

This project implements a simple handwriting digit detection model, focusing on binary classification of digits. The aim is to detect and classify handwritten digits into two classes: digit vs. non-digit (or another specified binary division).

Table of Contents

• Project Overview
• Dataset
• Model Architecture
• Installation
• Usage
• Results
• Future Work
• Contributing
• License

Project Overview

This project uses a machine learning model to detect handwritten digits, focusing specifically on binary classification. The binary classification can be adjusted based on the desired classes, for example:

• Digit 0 vs. Digit 1
• Digits 0-4 vs. Digits 5-9

The model is trained on images of handwritten digits and can be extended to more complex scenarios or modified for multi-class classification in the future.

Dataset

The project uses the MNIST dataset, a widely used dataset for digit recognition tasks. The MNIST dataset contains 60,000 training images and 10,000 testing images of handwritten digits, each of size 28x28 pixels.

Data Preprocessing

• Normalization: Pixel values are normalized to a range of 0-1.
• Binarization: The dataset is filtered or adjusted to convert it into a binary classification problem.

Model Architecture

The model is built using a simple neural network architecture. Here's a brief description:

• Input Layer: Takes an input vector of shape (400,).
• First Hidden Layer (layer1): A dense layer with 25 units and a sigmoid activation function.
• Second Hidden Layer (layer2): A dense layer with 15 units and a sigmoid activation function.
• Output Layer (layer3): A single neuron with a sigmoid activation function, suitable for binary classification.

Alternatively, convolutional neural network (CNN) layers can be added to improve accuracy on complex data.

Installation

To set up the project locally, follow these steps:

1. Clone this repository:

   git clone https://github.com/yourusername/handwriting-digit-detection.git