Classification of EEG Signal for Detecting Cybersickness through Time Domain Feature Extraction using NaÏve Bayes

• Published in: 2018 International Conference on Computer Engineering, Network and Intelligent Multimedia (CENIM) pp. 29 - 34
• Main Authors: Mawalid, Moch.Asyroful, Khoirunnisa, Alfi Zuhriya, Purnomo, Mauridhi Hery, Wibawa, Adhi Dharma
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.11.2018
• Subjects: Cybersickness; Electroencephalography; Electrooculography; Feature extraction; Games; K-Nearest Neighbor; Naive Bayes; Three-dimensional displays; Time Domain Feature Extraction; Time-domain analysis
• DOI: 10.1109/CENIM.2018.8711320

Recently the rapid developments in entertainment such as 3D movies and video games, causing the phenomenon of cybersickness to be a very serious topic among health experts. Cybersickness occurs when the human exposure in virtual environment so that it can cause negative effect like headache, fatigue, eyestrain and vomiting. It can disturb the physical and physiological of the human if it is not minimized properly. Many studies have been done to investigate cybersickness using several methods. One of the most common method is using Electroencephalograph (EEG). However, previously there were not many studies that explored time domain feature extraction in investigating cybersickness. In this paper, Nine healthy participants (7 male and 2 female) were measured using EEG during playing 3D video game. Time domain feature extraction, such as statistical features (e.g., mean, variation, standard deviation, number of peak) and power percentage band were implemented to recognize cybersickness. The frequency band alpha (\boldsymbol{\alpha}) and beta (\beta) was extracted for all channels. Then, we do the feature selection to improve the performance of cybersickness recognition using K-Nearest Neighbor and NaÏve Bayes classifier. We classified the result of feature extraction in order to investigate cybersickness symptoms or not. According to our research, the use of three feature extractions (i.e., variant, standard deviation, and number of peak) are the best feature for cybersickness recognition. The accuracy was 83,8% using Naive Bayes classifier. This result could improve the accuracy by 6% compared with the one that using five feature extractions.