Figure 5

Distributions of the features extracted from EEGNet for each cross-validation (CV) strategy when a window length of 10 s was used. Because a 60-s EEG epoch was used for data analysis, use of the 10-s window length resulted in 6 trials after data augmentation for both subject-wise CV (sCV) and trial-wise CV (tCV), and 21 trials for overlapped sCV (osCV) and overlapped tCV (otCV), respectively. Red circles, blue triangles, black rectangles represent the features extracted from PTSD patients, healthy controls (HCs), and an example participant (#109 PTSD patient), respectively. Empty and filled symbols represent training and test data, respectively. The features of PTSD patients and HCs are independently clustered with some overlap between the two groups for sCV and osCV while those are more independently and densely clustered with little overlap between the two groups for tCV and otCV than for sCV and osCV because tCV and otCV used trials augmented from a same participant for both training and test data simultaneously. otCV shows better separability than tCV due to more data leakage.