Seite - 31 - in Intelligent Environments 2019 - Workshop Proceedings of the 15th International Conference on Intelligent Environments

Figure 4. Multisensor machine learning workflow and reporting 4. Results from Experiments 4.1. Data Acquisition We deployed our multisensor sleep monitoring system in a bedroom and one subject used it during a week, in seven consecutive nights, from 8 to 14 of March. The present configuration of the monitoring system collects two samples per second of each environment sensor. In addition, we collected data from a Xiaomi Mi Band 2 wristband wore by the subject during the experiment, namely from the 3D accelerometer and the optical heart rate sensor. The data collected by the Fitbit wristband sensors during the same period allowed the definition of the sleep structure for each of the seven nights considered in this experiment. Table 3 summarizes the characterization of sleep for the referred period. The Fitbit app presents the sleep structure with estimated sleep stages. Figure 5 illustrates the structure obtained for the first night of the experimental study, 8 of March. Table 3. Sleep structure for the period of the experiment Week 8-14 March 2019 Sleep schedule Time asleep Time in each sleep stage Awake REM Light Deep Friday, 8 02:28 - 08:27 5hr 9min 50min 1hr 27min 2hr 35min 1hr 7min Saturday, 9 00:51 - 09:48 7hr 38min 1hr 19min 1hr 38min 4hr 21min 1hr 39min Sunday, 10 03:02 - 10:09 6hr 9min 58min 1hr 3min 4hr 1min 1hr 5min Monday, 11 00:50 - 07:00 5hr 19min 51min 55min 3hr 16min 1hr 8min Tuesday, 12 01:35 - 07:23 4h 52min 56min 1hr 12min 2hr 46min 54min Wednesday, 13 23:51 - 07:22 6hr 23min 1hr 8min 1hr 16min 3hr 55min 1hr 12min Thursday, 14 00:47 - 07:32 5hr 47min 58min 1hr 6min 3hr 33min 1hr 8min 4.2. Data Pre-processing A single dataset with the sensors data acquired from the 7 experiments was prepared, considering each entry of the dataset as summaries of one minute for all the sensors data. The summary values were calculated as the mean values for the considered time interval (1 minute). In the case of the 3D accelerometer, besides the mean values, the variance C.Gonçalvesetal. /MultisensorMonitoringSystem toEstablishCorrelations 31