MNIST Digit Classification with CNN

A Convolutional Neural Network implementation for handwritten digit recognition using the MNIST dataset. This project demonstrates the power of CNNs for image classification tasks.

Overview

This project implements a CNN model to classify handwritten digits (0-9) from the MNIST dataset. The model achieves high accuracy through convolutional layers, pooling, and dense layers, showcasing fundamental deep learning concepts for computer vision.

Features

• Data Visualization: Display sample MNIST digits with labels
• CNN Architecture: Custom convolutional neural network with:
  • Conv2D layer (32 filters, 3x3 kernel)
  • MaxPooling2D layer (2x2 pool size)
  • Flatten and Dense layers
  • Softmax output for 10-class classification
• Model Training: 20 epochs with validation monitoring
• Performance Evaluation: Test accuracy and loss metrics
• Predictions: Real-time digit prediction on test images
• Custom Image Testing: Load and predict custom digit images

Dataset

The MNIST dataset contains:

• Training set: 60,000 images
• Test set: 10,000 images
• Image size: 28x28 pixels (grayscale)
• Classes: 10 digits (0-9)

Model Architecture

Sequential Model:
├── Conv2D(32, (3,3), activation='relu') - Input: (28,28,1)
├── MaxPooling2D((2,2))
├── Flatten()
├── Dense(32, activation='relu')
└── Dense(10, activation='softmax')

Requirements

numpy
pandas
matplotlib
seaborn
tensorflow
keras
scikit-learn
pillow

Installation

1. Clone the repository:

git clone https://github.com/[username]/mnist-cnn-classifier.git
cd mnist-cnn-classifier

2. Install dependencies:

pip install numpy pandas matplotlib seaborn tensorflow scikit-learn pillow

Usage

1. Run the Jupyter notebook:

jupyter notebook Day5_Convolutional.ipynb

2. Key sections:
   • Data loading and preprocessing
   • Visualization of sample digits
   • Model creation and compilation
   • Training with performance tracking
   • Evaluation and predictions
   • Custom image testing

Results

• Training Time: Approximately 20 epochs
• Test Accuracy: High accuracy on MNIST test set
• Confusion Matrix: Detailed classification performance
• Custom Predictions: Supports external digit image classification

Key Concepts Demonstrated

• Convolutional Neural Networks: Feature extraction through convolution
• Data Preprocessing: Normalization and one-hot encoding
• Model Evaluation: Loss, accuracy, and confusion matrix analysis
• Image Processing: Custom image loading and preprocessing
• Deep Learning Pipeline: End-to-end ML workflow

File Structure

├── Day5_Convolutional.ipynb    # Main notebook with CNN implementation
├── README.md                   # Project documentation
└── requirements.txt            # Dependencies (optional)

Contributing

1. Fork the repository
2. Create a feature branch (git checkout -b feature/improvement)
3. Commit changes (git commit -am 'Add new feature')
4. Push to branch (git push origin feature/improvement)
5. Create a Pull Request

License

This project is open source and available under the MIT License.

Acknowledgments

• MNIST dataset from Yann LeCun's website
• TensorFlow/Keras for deep learning framework
• Matplotlib/Seaborn for data visualization

---

Note: This project is part of a deep learning series focusing on convolutional neural networks for computer vision tasks.