EEG-Based Cybersickness Classification Using Hjorth Parameters and Random Forest During 3D Gaming Exposure

Abstract

This study presents an efficient EEG-based framework for cybersickness classification utilizing Hjorth parameter features under a 3D immersive video game stimulus (Mirror Edge). Multimodal data acquisition was performed using a 14-channel Emotiv EEG system for objective measurement and the Simulator Sickness Questionnaire (SSQ) for subjective validation. The EEG signals were subjected to comprehensive preprocessing procedures, including band pass filtering and Independent Component Analysis (ICA) to eliminate artifacts. Then, using Discrete Wavelet Transform (DWT) to isolate theta,alpha, and beta bands. Hjorth parameters: activity, mobility, and complexity were subsequently extracted to capture the temporal dynamics of neural activity with low computational overhead. To mitigate feature redundancy and dimensionality, Correlation Feature Selection reduced the feature space from 126 to 9 salient features. Classification performance was evaluated using Random Forest, Support Vector Machine, and K-Nearest Neighbor. Experimental results indicate a consistent increase in SSQ scores across participants, with disorientation emerging as the predominant symptom. Random Forest achieved superior performance with an accuracy of 82%, outperforming K-NN (72.72%) and SVM (59.09%). Notably, feature reduction preserved Random Forest performance while enhancing alternative classifiers. These findings highlight the robustness and computational efficiency of the proposed approach, demonstrating its potential for real-time EEG-based cybersickness detection.