Power of Neural Networks in Digit Recognition

Overview

This project explores the power of neural networks in image classification, specifically for recognizing handwritten digits from the MNIST dataset. Starting with a simple dense neural network, we then move to Convolutional Neural Networks (CNNs) to leverage their advanced capability in processing spatial information, significantly improving accuracy.

Project Structure

1. Introduction: Background on neural networks and the motivation for using CNNs for image data.
2. Data Preprocessing: Loading, normalizing, and reshaping the data to prepare for neural network training.
3. Model Development:
   • Baseline model using a dense neural network.
   • Advanced model using a CNN for improved feature extraction and classification.
4. Training and Evaluation: Training models and comparing results, with the CNN achieving an impressive 0.99514 accuracy score on Kaggle.

Results

• Best Model: Convolutional Neural Network (CNN)
• Score: The final model achieved a score of 0.99514 on the Kaggle "Digit Recognizer" competition leaderboard.

Requirements

• Python 3.x
• Libraries: TensorFlow, Keras, NumPy, Matplotlib, etc.

Usage

1. Clone the repository.
2. Download the dataset from Kaggle.
3. Save the downloaded file in the same directory as the Jupyter Notebook.
4. Install the necessary libraries: pip install tensorflow keras numpy pandas matplotlib scikit-learn
5. Run the notebook to view model training and evaluation.

Acknowledgments

This project was created as an exercise in deep learning with a focus on image recognition tasks. Special thanks to the Kaggle community and MNIST dataset creators.