Automatic Eeg Classification Using Density Based Algorithms Dbscan and Denclue

Abstract

Electroencephalograph (EEG) is a commonly used method in neurological practice. Automatic classifiers (algorithms) highlight signal sections with interesting activity and assist an expert with record scoring. Algorithm K-means is one of the most commonly used methods for EEG inspection. In this paper, we propose/apply a method based on density-oriented algorithms DBSCAN and DENCLUE. DBSCAN and DENCLUE separate the nested clusters against K-means. All three algorithms were validated on a testing dataset and after that adapted for a real EEG records classification. 24 dimensions EEG feature space were classified into 5 classes (physiological, epileptic, EOG, electrode, and EMG artefact). Modified DBSCAN and DENCLUE create more than two homogeneous classes of the epileptic EEG data. The results offer an opportunity for the EEG scoring in clinical practice. The big advantage of the proposed algorithms is the high homogeneity of the epileptic class.

Detection Of Event-Related Potential Artifacts Of Oddball Paradigm By Unsupervised Machine Learning Algorithm

“…A lot of research has been done to remove artifacts but most methods require labeling the artifacts manually or requiring, additional hardware. For example, Electrooculography electrodes may require to place around the eyes or may necessitate data-sets containing a huge amount of data, and many more [15]. The involvement of humans to label artifacts in EEG data may be not desirable as it might be a tedious and time-consuming process [16].…”