MNIST Image Classification Project

This project is an implementation of a machine learning model for classifying handwritten digits from the MNIST dataset using TensorFlow/Keras. The model is a simple feedforward neural network that achieves high accuracy in recognizing the digits.

Getting Started

To get started with the project, follow the steps below:

Prerequisites

• Python 3.6 or higher
• TensorFlow 2.x
• NumPy
• Matplotlib (optional for visualization) You can install the required dependencies using the following command:

pip install tensorflow numpy matplotlib

Dataset

The MNIST dataset is already included in the TensorFlow/Keras library, so you don't need to download it separately.

Training the Model

To train the model, execute the train_model.py script:

python train_model.py

The script will load the dataset, preprocess the data, build the model, train it, and save the trained model weights to model_weights.h5. You can modify the hyperparameters, such as the number of epochs and batch size, in the script.

Evaluating the Model

To evaluate the model on the test set and print the accuracy, execute the processes.py script:

python processes.py

The script will load the saved model weights from model_weights.h5, evaluate the model on the test set, and print the test accuracy.

Visualizing Predictions

If you want to visualize the predictions for some test images, you can use the predict_images function in processes.py. By default, it prints the predictions for the first 10 test images. You can modify the num_images parameter to display predictions for a different number of images.

Model Architecture

The model used for this project is a simple feedforward neural network with three layers:

1. Input Layer: A fully connected (dense) layer with 128 units and ReLU activation function.

2. Hidden Layer: Another fully connected layer with 64 units and ReLU activation function.

3. Output Layer: The final layer with 10 units (corresponding to the 10 digits) and a softmax activation function.

Results

After training the model for 10 epochs, the accuracy on the test set is approximately 97%.

Further Improvements

• Experiment with different model architectures, such as convolutional neural networks (CNNs), to potentially improve accuracy.

• Try hyperparameter tuning to optimize the model's performance.

• Visualize the model's predictions on a larger set of test images to gain insights into its strengths and weaknesses.

License

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

Acknowledgments

• The MNIST dataset is provided by TensorFlow/Keras and can be accessed through the tensorflow.keras.datasets.mnist module.

• Parts of the code structure and ideas are inspired by various online tutorials and resources.