Applying Classification Algorithms to Identify Brain Activity Patterns

Abstract

AbstractThe paper examines the applicability of classification algorithms to identify brain activity patterns according to EEG data using the example of determining the state of eyes. In this area of research, the following main subareas can be distinguished: search for a features’ set that can be extracted from EEG data using a minimum number of electrodes, preprocessing techniques development for EEG data, classification techniques comparison and development of new ones, recommendations development for the selection of suitable classification algorithms. Particularly relevant is the question of how accurately classifiers can work with data coming in real time from the EEG neuroheadsets. The paper compares basic classification algorithms implemented in the Weka machine learning tool. For the experiment, six data sets were designed from the ‘‘EEG Motor Movement/Imagery’’ corpus. In the first stage of the experiment, 20 algorithms were investigated on one dataset. The best results were obtained by the IBk, RandomForest and RandomTree algorithms. In the second stage of the experiment, these three algorithms are compared on five additional data sets. The best result on four datasets was obtained for IBk, and one dataset best showed on the RandomForest. The study clearly demonstrated that using a simple data preprocessing procedure it is possible to obtain a classification model that works with an accuracy of 73%–93% even if one applies well-known machine learning algorithms. This preprocessing procedure consists of applying a bandpass filter and excluding outliers from data whose values are greater than three standard deviations from the median.KeywordsEEGClassification algorithmWekaIBk