Figure 1

Two data augmentation approaches based on a cropping method (overlap and non-overlap) and four cross-validation (CV) strategies (subject-wise CV, overlapped subject-wise CV, trial-wise CV, and overlapped trial-wise CV). EEG data of all subjects (N = 135) are augmented by cropping the whole EEG data independently. Five different window lengths were used for data cropping to see the impact of the number of trials (5, 10, 15, 20, and 60 s) with non-overlap and overlap strategies, respectively. For subject-wise CV, all augmented trials of a single participant are used as either training or test data together, whereas augmented trials are randomly divided into training and test regardless of the participant for trial-wise CV. Overlapped subject-wise CV and overlapped trial-wise CV use a same approach to subject-wise CV and trial-wise CV, respectively, except that trials augmented based on an overlapped window are used. Note that different colors represent different participants.