Murside Degirmenci ¹ Yılmaz Kemal Yüce ² Matjaž Perc ³ Yalcin Isler ^4*

• ¹ Kutahya Health Sciences University, Kutahya, Kütahya, Türkiye
• ² Alanya Alaaddin Keykubat University, Antalya, Antalya, Türkiye
• ³ University of Maribor, Maribor, Slovenia
• ⁴ Izmir Kâtip Çelebi University, Çigli, Türkiye

Motor Imagery (MI) Electroencephalography (EEG) signals are non-stationary and dynamic physiological signals which have low signal to noise ratio. Hence, it is difficult to obtain high classification accuracy. Although various machine learning methods have already proven useful to that effect, using many features and ineffective EEG channels often yields a complex structure of classifier algorithms. State-of-the-art studies were interested in improving classification performances with complex feature extraction and classification methods by neglecting detailed EEG channel and feature investigation in MI task prediction from EEGs. Here we investigate the effects of the statistically significant based feature selection method on four different feature domains -time-domain, frequency-domain, time-frequency domain, and non-linear domainand their two different combinations to reduce the number of features and to classify MI EEG features by comparing lower dimensional matrices with well-known machine learning algorithms.Our main goal is not to find the best classifier performance, but to perform feature and channel investigation in MI task classification. Therefore, the detailed investigation of the effect of EEG channel and feature is implemented using statistically significant features distribution among 22 EEG channels for each feature set, separately. We used BCI Competition IV Dataset IIa 1 Degirmenci et al.and 288 samples per person. A total of 1364 MI EEG features were analyzed in this study. We tested nine distinct classifiers: Decision tree, Discriminant analysis, Logistic regression, Naive Bayes, Support vector machine, k-Nearest neighbour, Ensemble learning, Neural networks, and Kernel approximation. Among all considered feature sets, the classifications performed using non-linear and combination feature set resulted in the maximum accuracy of 63.04% and 47.36% for binary and multiple MI task predictions, respectively. Ensemble learning classifier achieved the maximum accuracies in almost all feature sets for binary and multiple MI task classifications. Our research thus shows that the statistically significant based feature selection method significantly improves classification performance with fewer features in almost all feature sets, and that it enables detailed and effective EEG channel and feature investigation.