Figure 11:
Left: Arousal PDF over 30-second epochs for manual annotations in CFS (n = 145), MrOS (n = 580), WSC (n = 271) and model predictions over all test data (n = 996). The distribution is expected to be uniform as scoring of arousals is unrelated to this epoch, however annotations in CFS and MrOS exhibit a strong bias toward the central and later part of the 30-second epoch. Right: The arousal F1 score evaluated per second is lower in the first 10 seconds of each 30-second epoch due to this edge bias.