Towards Optimized Cybersickness Prediction for Computationally Constrained Standalone Virtual Reality Devices

Cybersickness, a spectrum of discomforts affecting 60-95% of virtual reality (VR) users, remains a major challenge for comfortable immersive experiences. While recent research using multimodal data and complex machine learning models for cybersickness prediction shows promise, their reliance on external computational resources (cloud servers, GPUs) renders them unsuitable for standalone VR devices (SVRs) due to their limited processing power and potential network lag, which can worsen cybersickness. To bridge this gap, we propose a novel approach that minimizes model complexity and training parameters through hyperparameter tuning, achieving comparable prediction accuracy and error rates with significantly reduced inference time. Our initial findings pave the way for future research on optimized cybersickness prediction models for computationally constrained SVR devices.