Quantitative EEG features and machine learning classifiers for eye-blink artifact detection: A comparative study

Abstract

Ocular artifact, namely eye-blink artifact, is an inevitable and one of the most destructive noises of EEG signals. Many solutions of detecting the eye-blink artifact were proposed. Different subsets of EEG features and Machine Learning (ML) classifiers were used for this purpose. But no comprehensive comparison of these features and ML classifiers was presented. This paper presents the comparison of twelve EEG features and five ML classifiers, commonly used in existing studies for the detection of eye-blink artifacts. An EEG dataset, containing 2958 epochs of eye-blink, non-eye-blink, and eye-blink-like (non-eye-blink) EEG activities, is used in this study. The performance of each feature and classifier has been measured using accuracy, precision, recall, and f1-score. Experimental results reveal that scalp topography is the most potential among the selected features in detecting eye-blink artifacts. The best performing classifier is Artificial Neural Network (ANN) among the five classifiers. The combination of scalp topography and ANN classifier performed as the most powerful feature-classifier combination. However, it is expected that the findings of this study will help the future researchers to select appropriate features and classifiers in building eye-blink artifact detection models.