Predicting Cybersickness Using Machine Learning and Demographic Data in Virtual Reality

The widespread adoption of virtual reality (VR) technologies is significantly hindered by the prevalence of cybersickness, a disruptive experience causing symptoms like nausea, dizziness, and disorientation. Traditional methodologies for predicting cybersickness predominantly depend on biomedical data. While effective, these methods often require invasive data collection techniques, which can be impractical and pose privacy concerns. Furthermore, existing research integrating demographic information typically does so in conjunction with biomedical or behavioral data, not as a standalone predictive tool. Addressing this gap, we investigated machine learning techniques that exclusively use demographic data to classify and predict the likelihood of cybersickness and its severity in VR environments. This method relies on noninvasive, easily accessible demographic information like age, gender, and previous VR exposure. It offers a more user-friendly and ethically sound approach to predicting cybersickness. The study explores the potential of demographic variables as standalone predictors through comprehensive data analysis, challenging the traditional reliance on biomedical metrics. We comprehensively presented the input data and statistical analysis and later carefully selected the widely used machine learning models from different classes, including k-nearest neighbors, Naive Bayes, Logistic Regression, Random Forest, and Support Vector Machine. We evaluated their performances and presented detailed results and limitations. The research findings indicate that demographic data can be used to predict the likelihood and severity of cybersickness. This research provides critical insights into future research directions, including data collection design and optimization suggestions. It opens new avenues for personalized and inclusive VR design, potentially reducing barriers to VR adoption and enhancing user comfort and safety.