Paper Key : IRJ************842
Author: Gautam Kumpatla
Date Published: 27 Oct 2023
Abstract
The use of data augmentation techniques, particularly those powered by Generative Adversarial Networks (GANs), has proven crucial for improving machine learning modelsperformance in a variety of fields. However, as these techniquesbecome more widely used, it is crucial to examine the possiblesources of bias that can be introduced. This work explores theimportant problem of bias in data augmentation, especially whenusing GANs, providing a thorough analysis of the difficultiesand consequences. The purpose of this research is to look atsituations when this bias is negligible and has no negative effectson performance. This carries out tests to measure bias in differentGAN-based DA configurations. The developed methodology isbased on the results to assess if GAN-based DA can effectivelyenhance a given dataset. Based on our attempts to minimize bias,the onward suggestion for mitigating it is the implementation ofGAN-based DA.
DOI LINK : 10.56726/IRJMETS45565 https://www.doi.org/10.56726/IRJMETS45565