Seite - 27 - in Intelligent Environments 2019 - Workshop Proceedings of the 15th International Conference on Intelligent Environments

economical, non-intrusive alternative, which can be used in the domestic environment and for long periods of time, however it is not as accurate as PSG [5]. This recent interest in sleep and concern for its measurement coupled with the recent development of technologies and growth in the consumer market for wearable health devices has allowed the emergence of multiple, non-intrusive, affordable sleep monitoring solutions that users can use at home, in some cases with very acceptable levels of precision [6] [7] [8]. So we can say that nowadays, with all this technology and quantity of devices available, it is relatively simple to obtain sleep data at home, unobtrusively and for long periods of time. However, what these solutions are not yet able to provide is a justification for the values obtained for the various sleep parameters, or what may have contributed to those particular outcomes. In fact, sleep is multi- dimensional and involves not only objective but also subjective parameters [4] [9], so it has not been easy to obtain a strict and concise definition of sleep quality [2]. On the other hand, sleep can also be affected by a particular set of contextual factors [3], such as environmental factors like temperature, humidity, luminosity, noise and poor air quality [10] [11]. Thus, it would be interesting and useful to be able to establish relationships between the obtained sleep data and relevant environmental factors for these results, thereby to help people to be aware and to properly interpret their sleep architecture and performance. This new opportunity for research led us to devise a multisensor system for monitoring the sleep environment to try to understand the effect that these factors have on the structure and quality of sleep. This issues were explored through an experimental study over a period of one week. The rest of the paper is organized as follows. Next section presents a brief characterization of human sleep and its main parameters, followed by a discussion of sleep monitoring systems that take into account the effect of environmental factors. We than continue describing the developed multisensor sleep monitoring system and the experimental study. Next section elaborates on the developed machine learning models and experimental results and finally, last section presents the conclusions of our findings and future work. 2. Related Work This section starts with a brief characterization of human sleep, followed by a discussion of some examples of sleep monitoring systems that take into account the effect of environmental factors. 2.1. Sleep Architecture Human sleep is characterized by two distinct phases, NREM (non-rapid eye movements) and REM (rapid eye movements), which alternate cyclically, as illustrated in Figure 1, making four to five sleep cycles in one night, lasting about 90 to 120 min each [3]. The NREM sleep phase, or slow sleep, is usually subdivided into three states or depth levels: N1 (onset of sleep), N2 (light sleep), and N3 (deep sleep) [12]. Often, the N3 state is also referred to as slow wave sleep (SWS), presenting regular breathing and heart rate and low muscle tone. The REM sleep phase presents irregular breathing and heart rate, null muscle tone and, of course, rapid eye movements. The N3 phase occurs essentially in the first third of sleep, whereas REM sleep occurs predominantly in the last third. C.Gonçalvesetal. /MultisensorMonitoringSystem toEstablishCorrelations 27