High-Frequency Cybersickness Prediction Using Deep Learning Techniques With Eye-Related Indices

Abstract

Cybersickness is a growing concern in the field of virtual reality (VR). It is characterized by symptoms, such as headache, sweating, disorientation, and nausea. These symptoms can considerably hinder the users' immersive experience in VR environments, leading to a pressing need for effective solutions to combat cybersickness. In this study, we aim to tackle cybersickness by presenting a novel high-frequency approach for detecting the timing at which users experience cybersickness. Our approach uses 1-, 5-, or 10-s time-series eye-related indices processed by deep learning algorithms to predict cybersickness severity. In five-fold cross-validation, we achieved 71.09% accuracy in classifying four classes of cybersickness severity when individuals were not distinguished. Furthermore, with individualized cross-validation, we achieved an accuracy of up to approximately 80%. Our approach outperforms other cybersickness prediction studies as it provides the highest frequency in predicting cybersickness. It is anticipated that our approach will be valuable not only for immediate evaluation by researchers investigating cybersickness mitigation but also for early detection and notification of users experiencing cybersickness symptoms. By predicting cybersickness, our approach has the potential to promote the future advancement of VR technology.