Ocular artifact from electroencephalogram – a comparative analysis of feature extraction, selection and classification

An electroencephalogram (EEG) is a record of signals that represent surface potentials varying whenever the brain performs any task and can be recorded by placing an arrangement of electrodes at the scalp of the brain. These recordings are often contaminated by unwanted movement near these electrodes, resulting in non-cerebral signals called artefacts. The presence of artefacts makes the study of EEG signals difficult. This work focuses on a comparative analysis of classification of ocular artefacts from EEG signal that mainly comprise of eye blinks. Various feature extraction, feature selection and classification techniques are used to compare the prediction performance of the system. Three different methods were used to extract features from the EEG recording done on eight subjects, performing two different tasks. Then the diagnostic performance of three feature selection and 30 classification methods were evaluated using 5-fold cross-validation. Performance of the system on various combinations has been calculated in terms of accuracy and results have been discussed. The maximum accuracy of 93.8% was yielded by classifiers: Kernel Naïve Bayes, Linear Support Vector Machine (SVM) and Ensemble Bagged Trees using wavelet-based features, principal component analysis as feature selection algorithm. By methodically assessing 360 feature-classifier combinations, this study is innovative and provides one of the most thorough benchmarks for ocular artefact identification with exceptional accuracy. It also has great potential for real-time EEG preprocessing in clinical and BCI applications.