Overview

In my MNIST dataset project, I leveraged a Convolutional Neural Network (CNN) to tackle the task of digit classification. Recognizing the importance of preventing overfitting, I adopted contemporary practices to ensure the model’s robustness. I implemented techniques such as dropout and resizing to enhance the model’s generalization capabilities while also applying data shuffling, a common practice to prevent the network from learning patterns specific to the dataset’s initial ordering. To further optimize the training process, I incorporated early stopping.

To thoroughly assess the model’s performance, I conducted 100 independent runs, each involving a unique shuffling of the dataset. This rigorous approach allowed me to achieve a remarkable accuracy rate of 99.3%, demonstrating the reliability and effectiveness of my CNN in accurately classifying handwritten digits.

Github Link: https://github.com/Alanl101/Mnist_Classification/tree/main