Virtual reality sickness detection: an approach based on physiological signals and machine learning

Virtual Reality (VR) is spreading to the general public but still has a major issue: VR sickness. To take it into consideration and minimize its occurrence, evaluation methods are required. The current methods are mainly based on subjective measurements and therefore have several drawbacks (e.g., non-continuous, intrusive). Physiological signals combined with Machine Learning (ML) methods seem an interesting approach to go beyond these limits. In this paper, we present a large-scale experimentation (103 participants) where physiological data (cardiac and electrodermal activities) and subjective data (perceived VR sickness) were gathered during 30-minute VR video game sessions. Using ML methods, models were trained to predict VR sickness level (based on the physiological data labeled with the subjective data). Results showed an explained variance up to 75% (in a regression approach) and an accuracy up to 91% (in a classification approach). Despite generalization issues, this method seems promising and valuable for a real time, automatic and continuous evaluation of VR sickness, based on physiological signals and ML models.