| Title |
Development of a Prediction Model for EEG-based Relaxation-arousal State of Users Experiencing a Virtual Reality Space |
| Authors |
김상희(Kim, Sang-Hee) ; 추승연(Choo, Seung-Yeon) |
| DOI |
https://doi.org/10.5659/JAIK.2022.38.11.107 |
| Keywords |
Virtual Reality Space; EEG; Relaxation-Arousal; Prediction Model; Machine learning |
| Abstract |
This study was carried out to develop a model that can predict a user's relaxation-arousal state by using electroencephalogram (EEG) data
and machine learning algorithms of users experiencing a virtual reality space. Specific ways were proposed to improve the prediction
accuracy of this model. Upon learning about this model, the prediction performance was compared while changing the hyperparameter
conditions of each model using supervised learning-based machine learning models suitable for the development of predictive models known
as the random forest, support vector machine, and artificial neural network algorithms. As a result, the random forest model had the highest
prediction accuracy when there were 300 trees, the support vector machine model when a sigmoid kernel was applied, and the artificial
neural network model when there were five hidden layers; these results confirmed that each optimal parameter condition could be met. Each
model was learned by applying the feature extraction method suggested in feature engineering to derive an improvement method in the
prediction performance of each model. The results revealed that when the frequency-specific statistics and filtering-based feature extraction
method was applied, the prediction performance improved in the random forest and artificial neural network models. Additionally, it was
shown that the machine learning models that could best predict the relaxation-arousal state from the EEG data of users experiencing a virtual
reality space was the artificial neural network model with five hidden layers applied with the frequency-specific statistics and filtering-based
feature processing method; its predictive accuracy was 70.21%. The results of this study could be useful basic data to implement an
automated system that evaluates the design drafts of a healing space by utilizing virtual reality and EEG data. |