VR video user experience evaluation under visual and temporal variations and machine learning–based cybersickness prediction

Abstract Virtual reality (VR) allows users to experience an engaging and interactive virtual environment. Such experience is often limited by cybersickness, a perceptual discomfort that varies among individuals. Understanding and modeling user responses to visual and temporal characteristics of VR content is essential for designing adaptive and user-centered systems. Despite its importance, the influence of variations in content encoding and playback on subjective perception of cybersickness remains underexplored. This study presents a user-centered evaluation of cybersickness in VR videos manipulated across three dimensions: quantization levels, resolutions, and playback delays. The study then conducted a subjective experiment with 34 participants who experienced these conditions via a head-mounted display (HMD). The user’s subjective responses were collected along with a simulator sickness questionnaire (SSQ). The subjective results indicate that cybersickness severity increases with higher compression levels, temporal discontinuities, and extreme resolutions, whereas moderate resolution (2.5K) and low compression (QP22) yield the most comfortable viewing experiences. Moreover, the study proposed a Machine Learning-based cybersickness levels prediction model. The subjective data were arranged and used to train multiple machine learning techniques, including ANN, SVM, Decision Tree, CatBoost, and XGBoost. A comparative evaluation among these techniques reveals that XGBoost consistently outperforms the other methods in terms of accuracy, F1-score, and AUROC, demonstrating strong discriminative capability and generalization performance. These findings highlight clear response patterns that can inform user modeling for adaptive VR systems, where content quality and playback conditions are dynamically adjusted to minimize discomfort. This work contributes to the broader goal of personalized, user-adaptive virtual environments that balance immersion and comfort through informed design.