4.2.1. EEG

Similar to our previous study [7], we extracted five EEG features: (1) ABP, (2) Normalized ABP (NABP), (3) differential entropy (DE), (4) rational asymmetry (RASM), and (5) differential asymmetry (DASM). As explained in Section 3.2, ABP is a PSD value that is produced at 10 Hz from each electrode and frequency band. NABP is a normalized value of ABP using the following equation:

$$\mathbf{x}' = \frac{\mathbf{x} - m\dot{\mathbf{m}}\,\left(\mathbf{x}\right)}{\max\left(\mathbf{x}\right) - m\dot{\mathbf{m}}\,\left(\mathbf{x}\right)},\tag{2}$$

where *x* is the original value, *x* is the normalized value, and *max*(*x*) and *min*(*x*) are the maximum value and the minimum value of the dataset, respectively.

To calculate the DE, RASM, and DASM features, we used the formulae proposed by Zheng et al. [4], which are replicated in Equations (3)–(5):

$$DE = \frac{1}{2} \log 2\pi \epsilon v^2,\tag{3}$$

$$
\Delta DASM = DE\_{(left)} - DE\_{(right)^\epsilon} \tag{4}
$$

$$RASM = DE\_{(left)} / DE\_{(right)}.\tag{5}$$

In Equation (3), *σ*<sup>2</sup> is the variance of the bandpass-filtered EEG data of each frequency band, and *π* and *e* are constants. Furthermore, as the equations show, DASM and RASM are based on DE and utilize the electrical current asymmetry between the electrodes of the EEG sensors. Therefore, in this study, we use the asymmetry between the electrodes and the values from each electrode at the same time to train classification models.

As explained in Section 4.1, we set the window size to 7 s for feature extraction. Therefore, 70 units of ABP data and 70 units of NABP data of each electrode and frequency band, and 1540 units of EEG data of each electrode were extracted from each participant's data. We calculated the average and the standard deviation for each 70 units of ABP and NABP data and used the results as features. In the calculation of DE, we used all 1540 units of EEG data. In more detail, we applied the bandpass filter on EEG data. Regarding the frequency range of filtering, we followed the frequency band ranges of our EEG sensor (see Section 3.2). Finally, RASM and DASM were calculated using the extracted DE values.

As a result of EEG feature extraction, we secured 40 features from ABP and NABP (five frequency bands times two electrodes times two summary statistics). Moreover, 10 features were extracted from DE (five frequency bands times two electrodes). Finally, as indicated by Equations (4) and (5), RASM and DASM utilize locational symmetry for each electrode and DE value. Thus, five features were secured from RASM and DASM (five frequency bands).
