**Abbreviations**


## **Appendix A**

In this appendix, we include any pre-processing details that were provided by the originators of the datasets used within this research.

#### *Appendix A.1. Driver Fatigue Data*

EEG was recorded using a 32-channel electrode cap, with two of the channels being reference channels linked to mastoid electrodes [38]. Scan 4.3 software of Neuroscan was used for preprocessing, with raw signals filtered by a 50 Hz notch filter and a 0.15 Hz to 45 Hz band pass filter in order to remove noise.

#### *Appendix A.2. Confused Students Data*

No known preprocessing information was provided by the originators.

#### *Appendix A.3. Alcoholism Data*

EEG correlates were sampled from 62 scalp electrodes and two EOG electrodes, at a sampling rate of 256 Hz [13]. Sampling started at 190 ms before onset of stimulus in order to record a pre-stimulus baseline, and EEG correlate durations provided in the dataset were 1 s in duration. Sensor values were provided in μV, resulting in a sequence

of 256 temporally organized values for each EEG channel. Trials with excessive eye or body movements (>73.3 μV) were rejected online. Only artifact free EEG segments were used to include eye blink artifacts. EEG electrodes were referenced to node site Cz during EEG measurement.

#### *Appendix A.4. PTSD Data*

For this dataset, there were 33 channels of EEG recorded, with two of the 33 channels being used for ground and reference, and at a sampling rate of 5000 Hz [34]. EEG preprocessing was performed in the proprietary software BrainVision Analyzer2. MRI gradient artifacts and cardio ballistic artifacts were removed using the template subtraction method. Then, the EEG was down-sampled to 250 Hz and filtered with a 40 Hz low-pass filter. Then, ICA was applied to remove residual cardioballistic artifacts as well as blink and saccade artifacts. Time periods of head motion were removed.

#### *Appendix A.5. Schizophrenia Data*

Vertical EOG (VEOG) and Horizontal EOG (HEOG) were also collected for the purpose of capturing eye movement and blinks [35]. Due to the size of the raw EEG signals, preprocessing was already performed on the dataset prior to its upload for public use. This preprocessing included re-referencing to the averaged mastoid channels, applying a 0.1 Hz high-pass filter, interpolation of outlier channels, and rejection of outlier components and outlier trials due to EEG artifacts using the FASTER artifact rejection method [54].
