The Identification of Significant Time-Domain Features for Wink-Based EEG Signals

Abstract

AbstractBrain-Computer Interface (BCI) is said to be a system that can measure and convert the brain activity into readable outputs. These outputs are said to be beneficial to the people who face physical challenges in carrying out their daily life as the outputs can be employed to control the BCI-based assistive device. Electroencephalography (EEG) is one of the electrophysiological monitoring techniques that record the brain’s electrical activity. Informative attributes can be extracted from the massive outputs of EEG signal and help in increasing the effectiveness of the BCI-based device. This study aims to discover the significant statistical time-domain features that can be used in the classification of the left wink, right wink and no wink utilising EEG signals. EMOTIV Insight was used as the EEG recording device to obtain the EEG signals triggered from the winking motion of the left and right wink. Six healthy subjects that ranged between 23 years old to 27 years old were involved in the wink-based EEG recordings. Nine statistical time-domain features were extracted, namely mean, median, standard deviation, variance, root-mean-square (RMS), minimum (Min), maximum (Max), skewness and kurtosis. The identification of the significant features is attained via a filter method known as information gain ratio. The ratio of training data to testing data was set to 70:30. The selected features for classification of winking is fed into various types of classifiers to observe the effect of this feature selection method on the performance of the classification, i.e. k-Nearest Neighbour (k-NN), Support Vector Machine (SVM), and Decision Tree. It was established from the present investigation that Standard Deviation, Variance and Min from channel AF4 were found to be significant. The classification accuracy (CA) for both train and test data with the filter feature selection method is observed to be comparably equal to the CA obtained from utilising all features. The findings from the study are non-trivial towards the realisation of a real-time BCI-based system.KeywordsElectroencephalogram (EEG)WinkingBrain-computer interface (BCI)Time-domain featuresMachine learningFeature selectionClassification