Figure 1.

Instance selection and normalization procedure. At least 7 out of the 14 days in the range of −28 to −14 relative to the ground truth day were retrievable. The mean (μ) and standard deviation (σ) from these days were used to normalize z-score wearable device metrics. We depict an example of a valid instance with its baseline period (−28 → −14) with retrievable data from 9 out of 14 nights (nights without retrievable data are indicated by a white cross). This instance is based on sleep summary features from the night before (night −1) and the night after (night 0) relative to the ground truth day.