This repository contains a PyTorch implementation of a Convolutional Neural Network (CNN) for classifying handwritten digits from the MNIST dataset. The model is trained using supervised learning and achieves high accuracy through convolutional feature extraction.

✅ Built with PyTorch – Uses PyTorch for defining and training the CNN.
✅ Uses MNIST Dataset – Standard dataset of handwritten digits.
✅ Convolutional Layers – Feature extraction using Conv2d.
✅ Max Pooling – Reduces spatial dimensions and prevents overfitting.
✅ Fully Connected Layers – Maps features to digit classes (0-9).
✅ Training & Evaluation – Tracks accuracy and loss metrics.
✅ Confusion Matrix – Evaluates misclassifications.

To set up the environment, install the required dependencies using:

pip install torch torchvision numpy pandas matplotlib scikit-learn

The MNIST dataset is automatically downloaded via torchvision.datasets. It consists of 60,000 training images and 10,000 test images, each 28×28 pixels in grayscale.

The CNN consists of:

1. Convolutional Layer 1: 6 filters (3×3 kernel), ReLU activation
2. Max Pooling 1: 2×2 pool size
3. Convolutional Layer 2: 16 filters (3×3 kernel), ReLU activation
4. Max Pooling 2: 2×2 pool size
5. Fully Connected Layer 1: 400 → 120 neurons (ReLU activation)
6. Fully Connected Layer 2: 120 → 84 neurons (ReLU activation)
7. Output Layer: 84 → 10 neurons (Log Softmax for classification)

Run the following script to train the model:

python cnnmnist.py

The model is evaluated based on:

• Accuracy: Percentage of correctly classified digits.
• Loss Curve: Tracks training and validation loss.
• Confusion Matrix: Highlights misclassifications.

A confusion matrix is generated to analyze misclassified digits.

📖 PyTorch Documentation
📖 Torchvision MNIST Dataset

📝 This project is licensed under the MIT License. Feel free to modify and use it!