Page - 30 - in Intelligent Environments 2019 - Workshop Proceedings of the 15th International Conference on Intelligent Environments

with an application developed for this purpose, allows the visualization and storage of the received data. Figure 3 illustrates the main interface of the referred android app. Figure 3. Android app to visualize and record ambient sensors data. In order to associate sleep structure with environment data, our multisensor monitoring system considers an activity wristband, Fitbit Charge 2, which uses a triaxial accelerometer and a heart rate sensor to estimate sleep structure. 3.1. Machine Learning profiling and Decision Support from Environment Attributes After the collection of sleep data, machine learning workflows are trained with the intention to detect the quality sleep and will be used in later stages as input to a decision support system to improve people wellbeing. The input attributes are found in the database and collected from a multiple number of sensors. The strategy is to find good decision models based on rule inference and extract attribute conditions learnt from those models and offer reports to the user and store the data in person specific profiles in the database, Figure 4. The attributes used are taken from environment and physical activity in order to add context to a data series which is used in common machine learning algorithms. The time series data are built with a data pre-processing stage following best practices from the CRISP-DM methodology [14]. Moving averages are built on top of original data to accommodate averages from data from the last 15 min and the last 30 min. From rule-based machine learning models, there is a creation of a list of attributes conditions, that can classify data instances. Such rules will be explored to perceive the environment characteristics that affect test subjects in their specific environment. It is expected that different test subjects and different environment produce different results. Reports to the users are made via direct notification to smartphone app. C.Gonçalvesetal. /MultisensorMonitoringSystem toEstablishCorrelations30