MNSIT Handwritten Digit Recognition using CNN

This research study uses the MNIST dataset to investigate how Convolutional Neural Networks (CNN) might be used to recognize handwritten numbers. From collecting data to evaluating models, it covers it all. We feed the preprocessed MNIST dataset into a CNN. It contains $\mathbf{6 0, 0 0 0}$ training images and $\mathbf{1 0, 0 0 0}$ test images. Using a range of hyperparameters, such as optimizers and learning rates the proposed CNN model’s layers such as convolutional, max pooling, and dense layers are fine-tuned. Additionally, the efficacy of batch normalization method is examined. This study measures the proposed model’s performance on test and training data using loss and accuracy metrics to ensure the model can be generalized. This study also proves that CNN works for number recognition and lays the groundwork for improvements in more advanced architectures, data augmentation, transfer learning, and integration with real-time applications.