Handwritten Number Recognition System

📌 Overview

The Handwritten Number Recognition System is a deep learning model built from scratch to classify handwritten digits using a custom neural network. It is trained on the MNIST dataset and utilizes NumPy-based implementation for forward and backward propagation.

🚀 Features

• Custom-built Neural Network: Implemented using NumPy without deep learning frameworks like TensorFlow or PyTorch.
• Forward & Backward Propagation: Includes ReLU and Tanh activation functions for optimization.
• Model Training & Evaluation: Supports training on both CPU and GPU.
• Manual & Automated Testing: Test the model using predefined test images or random inputs.
• Performance Metrics: Displays accuracy and prediction confidence.

🏗️ Tech Stack

• Python
• NumPy (for matrix computations)
• Matplotlib (for visualization)
• Pandas (for data handling)
• GPU Support (via CUDA for optimized training)

📂 Project Structure

Handwritten_Number_Recognition_System/
│── data/                     # Dataset storage
│── main.py                   # Loads trained model and tests data
│── model.npz                 # Saved model parameters
│── trainer.py                # Trains the neural network model
│── trainOnGPU.py             # Optimized training for GPU acceleration
│── README.md                 # Project documentation

📦 Installation & Setup

1. Clone the repository

   git clone https://github.com/Uni-Creator/Handwritten_Number_Recognition_System.git
   cd Handwritten_Number_Recognition_System

2. Install dependencies

   pip install numpy pandas matplotlib

3. Train the model (if needed)

   python trainer.py

4. Run the model for testing

   python main.py

📊 How It Works

1. The model loads pre-trained weights from model.npz or model.pth.
2. A test image is provided for prediction.
3. The model outputs a digit classification with confidence score.
4. The prediction is displayed along with the corresponding test image.

🛠️ Future Improvements

• Implement CNN-based architecture for improved accuracy.
• Add a web interface for user-uploaded handwritten digit classification.
• Support for different datasets beyond MNIST.

🤝 Contributing

Contributions are welcome! Feel free to open an issue or submit a pull request.

📄 License

This project is licensed under the Apache-2.0 license.

---