Multi-Modal Bio signal Transformer Based Deep Learning Approach for Mental Stress Detection

Abstract

Mental stress detection is a critical challenge in healthcare. The detrimental effects of stress on health and the societal expenses that accompany it are becoming more well-known because of initiatives like the World Health Organization and the Occupational Safety and Health Administration. There are over 20 million people living in Pakistan. Estimates indicate that up to 34% of Pakistanis may experience stress at some point in their lives, making it a common mental health issue in the country.

 This research developed a novel multi-modal bio-signal transformer based deep learning approach, leveraging transformer architecture, for accurate mental stress detection, a previously unaddressed challenge. The novel multi-modal deep learning approach, leveraging transformer architecture, for accurate mental stress detection. This approach utilizes a combination of bio signals, including electrocardiogram ECG, electroencephalogram EEG, electromyogram EMG, electrodermal activity EDA, TEMP and galvanic skin response GSR data, collected from participants under varying stress conditions. By employing deep learning techniques in conjunction with an encoder transformer and CNN, we can effectively classify these bio signals into binary stress states, indicating whether a participant is experiencing stress or not.

The model has achieved a remarkable 95% accuracy, surpassing previous state-of-the-art methods. The proposed method has the potential to improve the early detection and management of mental stress-related disorders and set a new benchmark in mental stress detection.