Handwritten Digit Recognition (MNIST)

This project implements a simple feedforward neural network (MLP) to recognize handwritten digits from the MNIST dataset.

It includes both a training script and a Tkinter-based GUI application where you can draw digits and get real-time predictions.

The project uses the MLP neural network library I created, which can be found at:

• Github: https://github.com/ThePhantom2307/MLP-Neural-Network
• PyPi: https://pypi.org/project/NeuralNetworkMLP/

Project Structure

├── main.py # Training script (loads MNIST, trains, saves model)
├── predict.py # GUI app for digit recognition
├── requirements.txt # Python dependencies
├── model.json # Saved trained model

Features

• Train a fully connected neural network (MLP) on MNIST.
• The trained model achieves ~97% accuracy (This accuracy is based on the Training Details).
• Save and load the model (model.json).
• GUI app where you can:
  • Draw digits on a canvas.
  • Predict the digit with probability distribution.
  • Clear and redraw easily.

Training Details

The model was trained on MNIST with the following setup:

• Input size: 784 (28x28 flattened images)
• Hidden layers: 512 and 256 neurons (ReLU activation)
• Output layer: 10 neurons (Softmax activation)
• Batch size: 32
• Epochs: 75
• Learning rate: 0.002

Results

• Validation accuracy: 96.94%
• Test accuracy: 96.93%

The trained model is saved to model.json.

Installation

1. Clone this repository:

git clone https://github.com/your-username/your-repo.git
cd your-repo

2. Install dependencies:

pip install -r requirements.txt

Training the Model

Run the training script:

python main.py

This will:

• Load the MNIST dataset.
• Train the neural network.
• Save the trained model to model.json.

Running the Digit Recognizer GUI

Once the model is trained (or model.json is present), launch the GUI:

python predict.py

Usage

• Draw a digit (0–9) on the canvas with your mouse.
• Click Predict to see the model’s prediction and probabilities.
• Click Clear to reset the canvas.

Requirements

The project depends on:

• numpy>=1.21.0
• keras>=2.4.0
• tensorflow>=2.4.0
• NeuralNetworkMLP>=1.1.2
• Pillow (for image processing in the GUI)
• tkinter (usually included with Python)

Notes

• The trained model (model.json) is generated after training and should be included if you want to run predictions without retraining.