Classification of arithmetic mental task performances using EEG and ECG signals

• Published in: The Journal of supercomputing 2023-09, Vol.79 (14), p.15535-15547
• Main Authors: Bergil, Erhan, Oral, Canan, Ergül, Engin Ufuk
• Format: Article
• Language: English
• Subjects: Accuracy; Arithmetic; Classification; Cognitive tasks; Compilers; Computer Science; Discriminant analysis; Electrocardiography; Electroencephalography; Feature extraction; Interpreters; Processor Architectures; Programming Languages; Signal classification; Support vector machines; Wavelet transforms
• ISSN: 0920-8542; 1573-0484
• DOI: 10.1007/s11227-023-05294-0

In this study, a classification is done for electroencephalogram (EEG) and electrocardiogram (ECG) records belong to arithmetic tasks with good and bad performances. Thus, it is aimed to evaluate the differentiation in mental arithmetic activity performance using EEG and ECG signals. First of all, EEG signals, taken from 36 subjects and labeled as good and bad performance according to the number of procedures performed within the same period, are divided into sections of 10 s. Sub-components are obtained using wavelet transform for segmented electroencephalogram and electrocardiogram recordings. The feature set is created by calculating the energy of the wavelet components from electroencephalogram recordings belong to 19 channels. The obtained feature set is classified by using logistic regression, support vector machines (SVM), linear discriminant analysis, and k-nearest neighborhood (k-NN) methods. The feature extraction process is repeated for electrocardiogram signals recorded during the arithmetic mental task, resulting in an extended feature set. The classification process in the expanded space is repeated using the same features. As a result of the analysis, it is observed that wavelet-based features are effective in determining mental activity performance. High accuracy classification is done by k-NN and SVM, respectively. For only EEG signals, the best classification result is obtained with k-NN with 97.22% accuracy, and for EEG and ECG signals are used together, the best result is obtained with k-NN with 99% accuracy. Features extracted from ECG signals have increased classification accuracy.