Random forest model predicts VR cybersickness with 86-user dataset

Virtual reality promises immersive experiences across gaming, education, healthcare, and training, but cybersickness remains a major barrier. Head-mounted displays can trigger nausea, dizziness, and disorientation, often forcing users to abandon sessions. In a new arXiv paper, Thiago Porcino, Erick Oliveira Rodrigues, Alexandre Silva, Esteban Clua, and Daniela Trevisan propose a machine-learning approach that predicts these symptoms before they fully manifest, giving developers and users a crucial window to intervene.

The team trained a random forest classifier, which outperformed 15 other machine-learning techniques across validation tests. To build their dataset, they designed a dedicated cybersickness profile questionnaire that captures a user's historical susceptibility and real-time responses to an immersive environment. They collected responses from 86 participants over multiple days, creating a rich dataset that blends individual physiological parameters with gameplay behavior.

Beyond predicting impending discomfort, their model also identifies the specific attributes—both environmental and personal—that contribute to stressful and uncomfortable VR experiences. This dual capability allows developers to dynamically adjust motion speed, field of view, or visual effects to reduce sickness risk. The research, presented at SeGAH 2020 and published on arXiv, marks a practical step toward personalized VR comfort systems.

The implications extend beyond gaming to medical rehabilitation, military simulators, and virtual classrooms where prolonged headset use is essential. By enabling real-time cybersickness prediction, this system could make VR more accessible and safer for millions of users, and opens the door for adaptive VR platforms that respond to each user's unique physiology.