Machine Learning Models for Mental Stress Classification based on Multimodal Biosignal Input

• Autor(a) principal: Justino, Maria Veríssimo Duarte
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/153937
• topic: Stress Detection; Biosignals; Occupational Health; Machine Learning; Multimodal Input; Domínio/Área Científica::Engenharia e Tecnologia::Outras Engenharias e Tecnologias

Mental stress is a largely prevalent condition directly or indirectly responsible for almost half of all work-related diseases. Work-Related Stress is the second most impactful occupational health problem in Europe, behind musculoskeletal diseases. When mental health is adequately handled, a worker’s well-being, performance, and productivity can be considerably improved. This thesis presents machine learning models to classify mental stress experienced by computer users using physiological signals including heart rate, acquired using a smart- watch; respiration, derived from a smartphone’s acc placed on the chest; and trapezius electromyography, using proprietary electromyography sensors. Two interactive proto- cols were implemented to collect data from 12 individuals. Time and frequency domain features were extracted from the heart rate and electromyography signals, and statistical and temporal features were extracted from the derived respiration signal. Three algorithms: Support Vector Machine, Random Forest, and K-Nearest-Neighbor were employed for mental stress classification. Different input modalities were tested for the machine learning models: one for each physiological signal and a multimodal one, combining all of them. Random Forest obtained the best mean accuracy (98.5%) for the respiration model whereas K-Nearest-Neighbor attained higher mean accuracies for the heart rate (89.0%) left, right and total electromyography (98.9%, 99.2%, and 99.3%) models. KNN algorithm was also able to achieve 100% mean accuracy for the multimodal model. A possible future approach would be to validate these models in real-time.