Machine-Deep-Ensemble Learning Model for Classifying Cybersickness Caused by Virtual Reality Immersion.

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Human-Computer Interaction

Background:

• Virtual reality (VR) immersion can induce cybersickness (CS), a condition affecting user experience.
• Accurate classification of CS is crucial for developing effective mitigation strategies.
• Physiological signals offer a potential objective measure for detecting CS.

Purpose of the Study:

• To develop and evaluate a machine-deep-ensemble learning model for classifying cybersickness (CS) induced by virtual reality (VR) immersion.
• To assess the model's ability to differentiate between neutral, non-CS, and CS states.
• To determine the efficacy of using physiological signals for CS detection.

Main Methods:

• Collected heart rate variability and respiratory signals from 20 subjects during a 5-minute VR video session.
• Developed a stacked ensemble model combining Support Vector Machine (SVM), k-nearest neighbor (KNN), Random Forest, and AdaBoost classifiers.
• Utilized a Convolutional Neural Network (CNN) for multiclass classification of CS states based on ensemble predictions.

Main Results:

• Individual models achieved high accuracies: SVM (94.23%), KNN (92.44%), Random Forest (93.20%), and AdaBoost (90.33%).
• The proposed ensemble model achieved a superior classification accuracy of 96.48% for the three CS states.
• The model demonstrated the capability to accurately detect neutral, non-CS, and CS states.

Conclusions:

• Machine-deep-ensemble learning models can effectively classify cybersickness (CS) from VR immersion using physiological data.
• The ensemble approach significantly enhances classification performance compared to individual machine learning models.
• This study validates the use of physiological signals and advanced computational models for objective CS detection.