Hag, Ala, Qazani, Mohammad Reza Chalak, and Asadi, Houshyar (2025) Quantifying Motion Sickness in Virtual Reality Using a Multimodal 1CNN–GRU–Attention Approach With GSR Data. IEEE Transactions on Intelligent Transportation Systems, 26 (12). pp. 22003-22014.

PDF (Published Version) - Published Version Restricted to Repository staff only

Abstract

Cybersickness remains a significant barrier to the widespread adoption of virtual reality (VR) and enhancing passenger comfort and safety in autonomous vehicles (AVs). Predicting and mitigating cybersickness is crucial for creating safer and more comfortable VR experiences. This study introduces a one-dimensional convolutional neural network-gated recurrent unit-attention (1CNN–GRU–Attention) model trained on galvanic skin response (GSR) data collected from 24 participants in VR under various weather conditions. Participants reported cybersickness severity on a 10-point scale every minute, providing real-time labels for evaluation. The model’s performance was compared with that of k-nearest neighbours (KNN) and support vector machine (SVM) classifiers in binary (two-class) and multiclass (four-class: none, low, acute, high) classification settings, both with and without class balancing using the synthetic minority oversampling technique (SMOTE). Evaluation metrics included accuracy (ACC), Matthews correlation coefficient (MCC), and unified performance metric (UPM), with performance assessed via 5-fold cross-validation and statistical t-tests. Results show that the 1CNN–GRU–Attention model significantly outperforms both SVM and KNN across all metrics. Applying SMOTE improved binary classification accuracy from 75.15% to 86.94% in binary classification and from 48.77% to 64.63% in multiclass classification, with notable improvements in MCC. These findings highlight the importance of balanced data and the effectiveness of the 1CNN–GRU–Attention model in assessing cybersickness severity, advancing physiological monitoring methods for VR, and facilitating broader VR adoption.

Actions (Repository Staff Only)

Item Control Page