Handwritten Digit Recognition API

This project is an implementation of a Handwritten Digit Recognition system using the MNIST dataset. The model currently uses a simple fully connected neural network, but I plan to upgrade it to Convolutional Neural Networks (CNNs) to improve performance and accuracy.

The final goal is to deploy this as a Flask API where users can upload their own digit images and receive predictions in real-time.

---

🚀 Features

• Train a neural network to recognize handwritten digits (0–9) from the MNIST dataset.
• Evaluate model accuracy and performance.
• Save and load trained models for reuse.
• Planned deployment as a Flask REST API for easy integration.

---

🛠️ Tech Stack

• Python
• TensorFlow / Keras
• Flask (planned deployment)
• NumPy, Matplotlib for data processing and visualization.

---

📈 Current Status

✅ Model trained using a basic neural network.
🔄 Plan to experiment with CNNs for better accuracy and generalization.
🚧 API implementation using Flask is in progress.

---

📂 Project Structure

.
├── MNIST_notebook.ipynb     # Jupyter notebook for model training and evaluation
├── saved_model/             # Directory to store trained model files
├── README.md                # Project documentation
└── requirements.txt         # Python dependencies

---

📌 Next Steps

1. Improve performance by switching from a dense neural net to a CNN architecture.
2. Implement Flask API for digit image upload and classification.
3. Enhance preprocessing to handle user-uploaded images in different formats.
4. Deploy API to a cloud platform for public use.

---

📦 Installation & Usage

1️⃣ Clone the repository

git clone https://github.com/your-username/handwritten-digit-recognition.git
cd handwritten-digit-recognition

2️⃣ Install dependencies

pip install -r requirements.txt

3️⃣ Train the model

jupyter notebook MNIST_notebook.ipynb

4️⃣ (Planned) Run Flask API

python app.py

---

📊 Example Predictions

Once deployed, the API will accept image uploads and return a JSON response like:

{
    "prediction": 7,
    "confidence": 0.98
}

---

📜 License

This project is licensed under the MIT License - see the LICENSE file for details.