The Effectiveness of Advance Deep Learning Architectures for Classification of Stress using Raw EEG Data

Understanding and detecting stress in today’s fast-paced world remains challenging. Previous research has explored various physiological signals, including EEG (Electroencephalography), to quantify stress. However, accurately assessing stress poses a significant challenge, particularly in preprocessing and feature extraction of EEG signals. Hence, this paper employs advanced deep learning algorithms, including ResNet18-1D, DenseNet-1D, and VGG16-1D, on raw EEG signals for automated stress classification. The study utilizes the publicly available EEG-based SAM40 dataset for stress classification and further validates the results on the EEG-based MAT dataset. The results demonstrate the efficacy of the ResNet18-1D model by achieving the highest of 98.73% and 97.24% on the SAM40 and MAT datasets, respectively. The findings pave the way for further exploration and utilization of raw EEG signals across diverse research domains beyond stress classification.