Figure 3

Mean and standard deviations of classification accuracies of different window lengths for two deep-learning models with respect to the cross-validation (CV) methods. Classification accuracies obtained using augmented data are significantly higher than that of the original data (60-s window length) in each CV method for the two deep-learning models (Bonferroni corrected p < 0.05), except for sCV method (20-s window length for CNN-13 and all window lengths for EEGNet) and for osCV method (all window lengths for EEGNet). n.s. means no significance, and the pairs without n.s. showed significant difference in terms of the classification performance.