Classification of Cognitive Stress with StressGAN-GP

This paper proposes an innovative Deep Learning (DL) architecture for classifying cognitive stress using a novel hybrid model, StressGAN-GP. The increasing prevalence of stress-related disorders in modern society necessitates accurate and efficient methods for stress detection and classification. This study aims to improve the model's generalization to various stress-inducing events by utilizing the capabilities of Generative Adversarial Networks (GANs) to synthesize stress-related patterns from limited labelled data. The suggested StressGAN-GP system blends the Gradient Penalty's interpretability with GANs' prowess in producing realistic stress characteristics. The proposed StressGAN-GP framework combines the strengths of GANs in generating realistic stress features with the interpretability of Gradient Penalty. The GAN component generates synthetic stress patterns, enriching the training dataset, while the GP component provides a probabilistic framework for stress classification, offering insights into model uncertainty. Test findings on an extensive dataset show that StressGAN-GP outperforms conventional classification models in terms of performance. The model accurately distinguishes between stressed and non-stressed individuals bearing an accuracy of 98.78%, showcasing its potential for real-world stress monitoring and intervention applications. This research contributes to advancing the field of affective computing and lays the foundation for more robust and interpretable models in the realm of cognitive stress classification.