From Physiological Measures to an Automatic Recognition System of Stress

Nicolas Martin (1) , Jean-Marc Diverrez (1)

1 Hypermedia - Domaine Hypermedia (IRT b<>com)

Evaluation of stress is mainly based on standardized scales. However, fill out questionnaires can be incompatible with several situations (e.g. during chirurgical intervention) and offers only subjective and punctual data. Physiological measures, which provided real-time and objective data, can be used to cope with these constrains. To be effective, physiological data need to be related to human feeling. One solution is to build an automatic recognition system of stress based on supervised machine learning. Thereby, to acquire physiological data, we built stressful situation in laboratory. From physiological data (respiratory, cardiac and electrodermal measurement) of 24 participants, we built a model that recognizes stress with an accuracy of 70%.

Mots clés: Stress; Physiological Measurement; Automatic Recognition; Machine Learning

Domaines: Psychologie; Machine Learning [stat.ML]

Nicolas Martin, Jean-Marc Diverrez. From Physiological Measures to an Automatic Recognition System of Stress. HCI International 2016, Jul 2016, Toronto, Canada. pp.139 - 176, ⟨10.1007/978-3-319-40542-1_27⟩. ⟨hal-01525673⟩