Development of a Classifier to Determine Factors Causing Cybersickness in Virtual Reality Environments.

Area of Science:

• Human-Computer Interaction
• Biomedical Engineering
• Virtual Reality Studies

Background:

• Cybersickness (CS) is a common issue in virtual reality (VR) experiences, impacting user comfort and immersion.
• Objective detection of CS is crucial for developing effective countermeasures and improving VR system design.
• Current methods often rely on subjective self-reports, which can be inconsistent.

Purpose of the Study:

• To develop and evaluate a classifier for detecting cybersickness (CS) occurrence in virtual reality (VR).
• To investigate the efficacy of combining biosignals and in-game parameters for CS detection.
• To establish a data-driven approach for identifying CS events during VR immersion.

Main Methods:

• Collected electrocardiographic, electrooculographic, respiratory, and skin conductivity data from 66 participants during a 10-minute VR experience.
• Recorded game parameters including avatar speed, acceleration, head movements, and collisions.
• Utilized machine learning classifiers (SVM, k-NN, neural networks) with simulator sickness questionnaire scores as ground truth for binary and ternary classification.

Main Results:

• Achieved a maximum classification accuracy of 82% for binary (CS vs. no CS) classification.
• Attained 56% accuracy for ternary (no CS, mild CS, severe CS) classification.
• Demonstrated that a combination of biosignals and game parameters can effectively indicate the occurrence of CS.

Conclusions:

• The study confirms that biosignals and game parameters are sufficient to detect the occurrence of cybersickness.
• Further research is needed to enhance binary classification accuracy for practical applications.
• Improved methods are required for accurately classifying the severity of cybersickness in real-time.