Enhancing Accuracy Through Data Augmentation using Variational Autoencoders in Machine Learning Techniques
Vaibhav Dubey, Bhavneet Kaur, Paurav Goel · 2024
This paper investigates the impact of data augmentation on the performance of a simple Deep Neural Network (DNN) architecture, comprising a sequence of fully connected layers interspersed with non-linear activation functions and regularization techniques. The model was trained over 2000 epochs, with and without data augmentation. Rectified Linear Unit (ReLU) activation functions are applied after each layer, except the last, to introduce non-linearity, while dropout regularization is used to enhance model robustness and prevent overfitting. Experimental results reveal that data augmentation significantly enhances model performance. The registered training loss decreased and training accuracy improved significantly with data augmentation technique. The same results are also observed in validation phase. These improvements suggest that data augmentation aids the model in learning more effectively by introducing variability in the training data, enhancing its generalization capability and robustness to variations in unseen datasets.