Cross-Game Generalization and Context Dependence of Cybersickness Prediction Using Consumer-Grade HMD Motion Data

Cybersickness remains a persistent barrier to sustained virtual reality (VR) use. Consumer-grade head-mounted displays (HMDs) provide motion data that can support symptom prediction, but most prior models have been evaluated within a single game or data environment, leaving cross-game generalization uncertain. This study reanalyzes HMD motion trajectories and Simulator Sickness Questionnaire (SSQ) responses from 60 complete participants across four commercial VR games in the publicly available Questset dataset. Fifteen session-level motion features were extracted, symptom change was represented as ΔSSQ, and Ridge Regression and Random Forest were evaluated under within-game and cross-game conditions. Cross-game error increased most clearly when Beat Saber was the target game under both models, whereas other transfer directions showed smaller penalties. Several feature–symptom correlation directions also changed across games, although none of the 60 exploratory correlations remained significant after False Discovery Rate (FDR) correction. These results suggest that HMD-based cybersickness prediction may be context-dependent and should not be assumed to transfer automatically to unseen VR content; larger within-subject, multi-game studies are needed to confirm this pattern.