Applied Sciences, Journal Year: 2024, Volume and Issue: 14(23), P. 10818 - 10818

Published: Nov. 22, 2024

This study presents a novel approach using Wasserstein Generative Adversarial Networks with Gradient Penalty (WGAN-GP) to generate synthetic electroencephalography (EEG) and electrocardiogram (ECG) waveforms. The EEG data represent concentration relaxation mental states, while the ECG correspond normal abnormal states. By addressing challenges of limited biophysical data, including privacy concerns restricted volunteer availability, our model generates realistic waveforms learned from real data. Combining datasets improved classification accuracy 92% 98.45%, highlighting benefits dataset augmentation for machine learning performance. WGAN-GP achieved 96.84% representing states optimal when classified fusion convolutional neural networks (CNNs). A 50% combination yielded highest 98.48%. For signals, consisted 60-s recordings across four channels (TP9, AF7, AF8, TP10) individuals, providing approximately 15,000 points per subject state. contained 1200 samples, each comprising 140 points, outperformed basic generative adversarial network (GAN) in generating reliable support vector (SVM) classifier an 98% 95.8% Synthetic random forest (RF) classifier’s 97% alone 98.40% combined Statistical significance was assessed Wilcoxon signed-rank test, demonstrating robustness model. Techniques such as discrete wavelet transform, downsampling, upsampling were employed enhance quality. method shows significant potential scarcity advancing applications assistive technologies, human-robot interaction, health monitoring, among other medical applications.

Language: Английский