Accurate medical diagnosis is vital for saving lives through early identification of illnesses. While machine learning (ML) models hold promise in medical research, obtaining high-quality datasets for training poses challenges in terms of time, money, and privacy concerns. This study explores generative models for medical data augmentation, utilizing real-world datasets. Synthetic data is generated to augment existing datasets, enabling the training of ML and AI models. Performance evaluation on a reserved segment of real-world data reveals that models trained on real and augmented data yield optimal results. Notably, XGBoost demonstrates consistent and promising performance across real, synthetic, and augmented datasets for liver disease prediction. In contrast, support vector machines exhibit the highest accuracy, achieving 98% on augmented and 96% accuracy on fully synthetic kidney disease data. These findings highlight the potential of synthetic data in sensitive medical synthesis during model training and offer insights for selecting models in medical prediction tasks.
Utilizing Generative Adversarial Networks for Medical Data Synthesis and Augmentation to Enhance Model Training
Smart Innovation, Systems and Technologies
Congress on Control, Robotics, and Mechatronics ; 2024 ; Warangal, India February 03, 2024 - February 04, 2024
Proceedings of the Second Congress on Control, Robotics, and Mechatronics ; Chapter : 8 ; 85-98
2024-10-31
14 pages
Article/Chapter (Book)
Electronic Resource
English
British Library Conference Proceedings | 2022
|