EVALUATION
In order to find the optimum minimum length for the
intervals of non-movement, we defined six different interval
thresholds (300, 360, 480, 600, 720 and 900 seconds) and evaluated their performance in regard to sleep detection, using the
PSG dataset as ground truth. We take the accuracy for detecting
sleep and wake phases into regard, and use the precision and
recall to investigate further differences between the individual
parameters’ performances. We observe that the mean accuracy
is best for the 600 interval which is why we choose this
interval to determine immobile segments. Additionally, we set
the standard deviation (STD) threshold to 6, as derived by
experimental evaluation of different thresholds.