Year: 2024
Type: conference-paper
Abstract: Stress has emerged and continues to be a regular obstacle in people's lives. When left ignored and untreated, it can lead to many health complications, including an increased risk of death. In this study, we propose a foundation model approach for stress detection without the need to train the model from scratch. Specifically, we utilise the foundation model "Neuro-GPT", which was trained on a large open dataset (TUH EEG) with 20,000 EEG recordings. We fine-tune the model for stress detection and evaluate it on a 40-subject open stress dataset. The evaluation results with a fine-tuned Neuro-GPT are promising with an average accuracy of 74.4% in quantifying "low-stress" and "high-stress". We also conducted experiments to compare the foundation model approach with traditional machine learning methods and highlight several observations for future research in this direction.
Authors: Catherine Lloyd, Loic Lorente Lemoine, Reiyan Al-Shaikh, Kim Tien Ly, Hakan Kayan, Charith Perera, Nhat Pham
Institutions: Cardiff University, University of Oxford
Language: English
Topic: Emotion and Mood Recognition
Subfield: Experimental and Cognitive Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: bronze
OpenAlex ID: W4405014290