EEG Eye State Identification based on Statistical Feature and Common Spatial Pattern Filter

• Main Authors: Woon, W.C., Yahya, N., Badruddin, N.
• Format: Conference or Workshop Item
• Published: Institute of Electrical and Electronics Engineers Inc. 2019
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85075633150&doi=10.1109%2fSCORED.2019.8896270&partnerID=40&md5=741df143ed27627f403d55ec2c98023d; http://eprints.utp.edu.my/24906/

EEG signal is one of the main sources for implementation of Brain-Computer Interface (BCI) technology. The BCI is a non-muscle communication link between brain and external device, which commonly designed to enable patients with neurological condition to interact with others using their brain signals. In this work, we investigated the classification of EEG eye state data using statistical and CSP filter technique. Statistical feature has been applied in EEG signal classifications of eye-close and eye-open conditions but the accuracy is reported to be less than 78. CSP filter is a well-known method for classification of motor imagery EEG in the BCI field but when applied for EEG eye state classification, it only gives accuracy similar to statistical feature, that is less than 78. These indicate that both methods give good discrimination of the eye state condition but on it own, will not be sufficient to produce good classification accuracy. Hence, this work aims to develop an algorithm using statistical-CSP feature for eye state classification from EEG signal. This is taking advantage on the discriminative feature provided by both methods, statistical and CSP filter, which is expected to increase the accuracy of the eye state classification algorithm. The process of developing the EEG eye state classification algorithm, includes data extraction, pre-processing, data normalization, feature extraction, feature selection and classification are detailed out in this paper. Number of selected electrodes are divided into 3 groups, with 2 groups having a set of 7 different electrodes and 1 group that combined both sets of 7 electrodes giving a total of 14 electrodes. Using ten-fold cross validation, the highest accuracy of statistical feature is at 54.3 and the highest accuracy of CSP feature is at 72.3 generated using fine Gaussian SVM classifier. Result from this work has shown that combining both statistical and CSP features from 7 electrodes of Group I has shown to result in good accuracy of 99.92. © 2019 IEEE.