Investigating Personalization Techniques for Improved Cybersickness Prediction in Virtual Reality Environments.

Area of Science:

• Virtual Reality
• Human-Computer Interaction
• Biomedical Engineering

Background:

• Cybersickness is a growing concern in virtual reality (VR).
• Predictive models often lack personalization, ignoring individual susceptibility factors like age and gender.
• Current models do not account for individual differences in predicting VR-induced motion sickness.

Purpose of the Study:

• To investigate the impact of personalization techniques on cybersickness prediction accuracy.
• To develop and evaluate personalized models that outperform non-personalized approaches.
• To enhance the inclusivity and user experience in virtual reality.

Main Methods:

• Utilized an open-source cybersickness dataset.
• Explored four personalization techniques: data grouping, transfer learning, early shaping, and sample weighing.
• Employed Deep Temporal Convolutional Neural Network (DeepTCN) for prediction modeling.

Main Results:

• Personalized models demonstrated significant improvements in prediction accuracy.
• The DeepTCN model with early shaping achieved a 69.7% reduction in Root Mean Square Error (RMSE) compared to its non-personalized counterpart.
• Empirical results confirm the benefits of personalization techniques in enhancing cybersickness prediction.

Conclusions:

• Personalization techniques are crucial for improving the accuracy of cybersickness prediction models.
• Tailored models can lead to more effective and inclusive virtual reality experiences.
• Findings support the development of personalized cybersickness reduction strategies.