Neural network training for handwritten digit classification (0–9).

The aim of this project is to train Deep Neural Networks (DNNs) for the classification of handwritten digits (0–9).

• The MNIST dataset was loaded using TensorFlow.
• Images of size 28x28 pixels were flattened into vectors of length 784, while labels were encoded using one-hot encoding.
• The initial network architecture included:
  • An input layer with 784 neurons
  • Two hidden layers with 256 neurons each, using the tanh activation function
  • An output layer with 10 neurons, using softmax activation
• Training used Stochastic Gradient Descent (SGD) with a learning rate of 0.001.
• The model was trained for 10 epochs.
• Performance optimization was explored by varying:
  • Number of layers
  • Number of neurons per layer
  • Number of epochs
  • Learning rate
  • Activation function
• A dropout rate of 0.3 was applied to the hidden layers, although this did not lead to accuracy improvements.
• For each model, both accuracy and loss were recorded during training and validation.

The final model achieved an accuracy of over 95% on the test dataset, indicating that the neural network is highly effective for handwritten digit recognition.