This work presents a novel approach using Wasserstein Generative Adversarial Networks with Gradient Penalty (WGAN-GP) to generate synthetic EEG waves corresponding to concentration and relaxation mental states. By addressing the challenge of limited biophysical data, our model produces realistic synthetic brainwaves learned from real data, which is vital for studies restricted by volunteer availability and privacy concerns. Combining real and synthetic data improved classification accuracy from 92% to 98.45%, highlighting the benefit of expanding the dataset for better machine learning model performance. The WGAN-GP model generated synthetic EEG data with 96.84% accuracy for relaxation and optimum accuracy for concentration, as classified by a Convolutional Neural Network (CNN). Additionally, 50% synthetic data combined with the original dataset achieved the highest accuracy (98.48%). A comparison was made between a basic GAN and the WGAN-GP model, demonstrated that the WGAN-GP outperformed the basic GAN in this study. Techniques like discrete wavelet transform, downsampling, and upsampling were employed. This method shows potential in addressing EEG data scarcity and enhancing human-robot interaction in assistive technologies and mental health monitoring.