Feature extraction method of EEG based on wavelet packet reconstruction and deep learning model of VR motion sickness feature classification and prediction

Abstract

Abstract The increased utilization of virtual reality (VR) technology has raised concerns regarding VR-induced motion sickness, an adverse condition linked to discomfort and nausea within simulated environments. Our approach commenced with an extensive analysis of EEG data and subjective feedback obtained from users immersed in VR environments. This data served to train a sophisticated deep learning model, employing an enhanced short-term memory network (GRU), aimed at identifying patterns and features associated with motion sickness. Following this, comprehensive data pre-processing and feature engineering were conducted to ensure the accuracy and suitability of the input data. Subsequently, a deep learning model was trained utilizing both supervised and unsupervised learning techniques, enabling the classification and prediction of motion sickness severity. Rigorous training and validation procedures were employed across multiple datasets to ascertain the model's robustness and performance under diverse scenarios. The research outcomes affirm the precision of our deep learning model in accurately classifying and forecasting the degree of motion sickness induced by virtual reality. The classification task achieved an accuracy rate of 84.9%, surpassing correlation and error indices of existing models. Consequently, this model exhibits superior capabilities in diagnosing and predicting motion sickness, thereby offering crucial support for enhancing the quality of the virtual reality experience and furthering advancements in VR technology.