Eye state classification from electroencephalography recordings using machine learning algorithms

Abstract

Background and Objectives: Current developments in electroencephalography (EEG) foster medical and nonmedical applications outside the hospitals. For example, continuous monitoring of mental and cognitive states can contribute to avoid critical and potentially dangerous situations in daily life. An important prerequisite for successful EEG at home is a real-time classification of mental states. In this article, we compare different machine learning algorithms for the classification of eye states based on EEG recordings. Materials and Methods: We tested 23 machine learning algorithms from the Waikato Environment for Knowledge Analysis toolkit. Each classifier was analyzed on four different datasets, since two separate approaches - called sample-wise and segment-wise - in combination with raw and filtered data were applied. These datasets were recorded for 27 volunteers. The different approaches are compared in terms of accuracy, complexity, training time, and classification time. Results: Ten out of 23 classifiers fulfilled the determined requirements of high classification accuracy and short time of classification and can be denoted as applicable for real-time EEG eye state classification. Conclusions: We found that it is possible to predict eye states using EEG recordings with an accuracy from about 96% to over 99% in a real-time system. On the other hand, we found no best, universal method of classifying EEG eye states in all volunteers. Therefore, we conclude that the best algorithm should be chosen individually, using the optimal classification accuracy in combination with time of classification as the criterion.

A novel method for EEG based automated eyes state classification using recurrence plots and machine learning approach

“…The classification error rate for a study 31 is very large, that is, 27.39%, as compared to 2.73% as computed for LR in our work. Very good performance in terms of classification accuracy has been achieved by the studies as, 97.5%, 34 99.8%, 39 97%-99%, 38 97.4%, 32 97.3%, 30 88.2%, 36 and 84.05%, 33 but all of them have explored very smaller datasets, that is, data from only a single subject has been used in the studies 30,[32][33][34]36,39 and data from 27 subjects in a study. 38 For the studies 29,37 based upon 11 subjects and 50 subjects, respectively, no classification accuracies have been computed by them.…”

“…Saghafi et al 36 proposed an algorithm that could perform an EEG eyes state change detection in less than 2 s that was the significant improvement over the Rösler and Suendermann 30 work; Hussain et al 37 found multi‐scale permutation entropy as the useful parameter in differentiating the EEG signals for the binary eyes states‐closed and opened; Piatek et al 38 performed three‐states (eyes open, eyes close, and blinking) classification using 23 different machine learning methods and obtained good performance in terms of classification accuracy, training time, classification time, and complexity; Zhou et al 39 developed an efficient eyes state classification system that gave excellent performance results in terms of classification speed and accuracy; Karamacoska et al 40 adopted the principal component analysis for the EEG and event‐related potential (ERP) components to examine the resting state conditions in relation to performing an auditory go and no‐go tasks.…”

“…Tsuji et al 28 proposed probabilistic neural network based on log-linearized Gaussian mixture model (LLGMM) for distinguishing the EEG patterns for eyes opening and closing scenarios; Li 29 adopted the spectral analysis approach and computed the differences in the absolute band powers for eyes open, close and, attention states; Rösler and Suendermann 30 built an EEG eye state corpus using 14 channel emotive headset and applied 42 different machine learning algorithms and statistical methods for eye state prediction; Wang et al 31 used an incremental attribute learning approach based on neural networks for performing an EEG based eyes state classification; Hamilton et al 32 developed a system using an ensemble of learners and improved the performance over the work done by Rösler and Suendermann 30 in terms of classification speed and accuracy; Sabanc𝚤and Koklu 33 experimented that k-nearest neighbors (KNN) algorithm outperformed multilayer perceptron (MLP) network in terms of classification success rates; Reddy and Behera 34 employed deep neural architectures and obtained improved performance over the work done by Hamilton et al 32 and Rösler and Suendermann 30 in terms of classification accuracy and convergence time; activity recognition system developed by Kaur et al 35 suggested that statistical features such as root mean square and standard deviation and extracted from gamma frequency band act as the most distinguishing features for classifying eyes open and closed states. Saghafi et al 36 proposed an algorithm that could perform an EEG eyes state change detection in less than 2 s that was the significant improvement over the Rösler and Suendermann 30 work; Hussain et al 37 found multi-scale permutation entropy as the useful parameter in differentiating the EEG signals for the binary eyes states-closed and opened; Piatek et al 38 39 developed an efficient eyes state classification system that gave excellent performance results in terms of classification speed and accuracy; Karamacoska et al 40 adopted the principal component analysis for the EEG and event-related potential (ERP) components to examine the resting state conditions in relation to performing an auditory go and no-go tasks. Most of the above studies are based upon the traditional signal-processing techniques combined with machine learning approach.…”

“…Most of the above summarized studies [30][31][32][33][34]36,39 have been validated on a publicly available University of California, Irvine (UCI) eyes state corpus that contains the EEG data for only a single subject. For the experiments performed in the available studies, 28,38 data from 5 and 27 subjects, respectively, have been collected to validate the performance of the proposed models. So it is very important to generalize the analysis results on huge datasets for the deployment of the models in daily applications.…”

Effect of time windows in LSTM networks for EEG-based BCIs

“…In our case, no pre-selection of the signal was done when applying the proposed methodology. Other authors do select the best segments to build their dataset, which is not applicable in real-time systems; for example, Piątek et al ( 2018 ) obtained an accuracy of over 96% with different classification algorithms with a time window of 10 s.…”