A Comparative Analysis of Data Augmentation Techniques for Human Disease Prediction from Nail Images

Using vast amounts of data can enhance the effectiveness of machine learning algorithms and prevent overfitting issues. It might be difficult and time-consuming to gather a lot of training data for a disease prediction model in the health industry. Without the requirement to gather additional data, data augmentation approaches can broaden the range of data related to training, to overcome this problem. Various methods for enhancing images using deep learning, include color alteration, Neural Style Transfer (NST), image rotation, image cropping, PCA color augmentation, generative adversarial networks (GANs), noise injection, image rotation as well as flipping techniques were used in this study to create enhanced nail image datasets. Modern transfer learning techniques were used to assess the success rate of data expansion approaches, and then the results demonstrated that the supplemented dataset produced by NST as well as GAN methods had higher precision than the primary dataset. The blended approach of artificial intelligence, color, as well as orientation augmentation performed the best across all datasets.