Research Information System
Journal of the Faculty of Engineering and Architecture of Gazi University, vol.39, no.3, pp.1597-1609, 2024 (SCI-Expanded)
In Motor Imagery (MI) task classification, finger movement classification is less frequently analyzed due to their noisy signal characteristic that makes the difficult discrimination of finger movements. This study proposes a method investigating statistically significant time-domain features for classification of finger movement and no imaginary task condition (NoMT). Unlike existing MI task classification studies, which always assume that the MI task is fulfilled and ignore the NoMT condition, this study also evaluated the NoMT condition for a more realistic prosthesis design. The 24 different time-domain features are extracted from 19 EEG channels of 8 subjects. The statistically significance (ANOVA) based feature selection method is used to define significant time-domain features. Eight distinct typical classifiers with different parameters were computed to classify 6 groups using 5-fold cross-validation. In subject-independent and subject-dependent analysis, the highest test accuracy value was calculated as %35,9 and %33.3-%57.5 using ANOVA-selected time-domain features and Support Vector Machine (SVM) algorithm, respectively. As a result of study, the performances of the subject-dependent classifications are higher than the performance of the subject-independent classifications. These highest subject-dependent results are encouraging for the design of the accurate EEG-based BCI systems in future studies on the design of personalized hand prostheses.