Deep neural networks are powerful but need lots of good data to work well. When there isn't enough data, "Data augmentation" can help by making more data, but it might not be great quality. Overfitting is a common problem when training deep learning models, caused by imbalanced data and starting parameters. To fix this, Dropout regularization is used. In this study, Data augmentation and Dropout are used on the MNIST dataset to improve accuracy. The results show that using Dropout can help detect if a model is underfitting or overfitting, leading to better accuracy at first but then dropping off.