Data Augmentation Using GAN for Parkinson’s Disease Prediction

The disease of Parkinson is a gradual neurodegenerative disorder affecting approximately one million U.S. citizens with nearly sixty thousand new annual clinical health diagnoses [1]. Analysis of voice samples was used to detect Parkinson’s disease early (PD) as an efficient tool. The use of deep learning is dependent on the number of samples marked out, which limits the use of deep learning in the smaller sample environment. In this paper, we suggest an approach based on a GAN combined with a deep neural network (DNN). The initial samples were first divided into training and a test range. The GAN learned to generate synthetic sample data to expand the dataset. Last, the synthetic samples are prepared for the DNN classifier. Finally, the classifier testing conducted with the test set, and the indicators confirmed the efficacy of the small sample classification method. Experimental tests have shown greater precision than conventional approaches in the proposed plan. While the classification process appeared to be improved by traditional data increase, an increase of 11:68% was achieved by incorporating GAN-based additions. Moreover, even higher efficiencies can be obtained by combining conventional with GAN-based augmentation schemes.