Abstract

Introduction: As the demand for resilient deep learning models in medical imaging grows, the importance of data augmentation becomes increasingly evident. This study provides a comprehensive comparative analysis of advanced data augmentation techniques applied to a skin melanoma dataset.

Method: Through rigorous experimentation and evaluation, we aim to demonstrate the effectiveness of these methods in enhancing model generalization and reducing overfitting. The performance metrics selected, including accuracy, precision, recall, and F1-score, were chosen based on their relevance to assessing model robustness and generalization capability.

Results: Our findings offer valuable insights into the strengths and limitations of various advanced data augmentation techniques, aiding researchers and practitioners in making informed choices tailored to their specific applications.

Conclusion: Additionally, this study reviews several patents related to data augmentation in deep learning, ensuring that our approach is grounded in current technological advancements.

Keywords: Augmentation, small dataset, mixup, cutmix, random erasing, accuracy