Smart Home and Assisted Living – Evaluation of the medium stage demonstrator

posted in: Blog | 0

Following the successfull evaluation of the early SHAL demonstrator, development went on towards implementation of the medium stage. The evaluation discussed here revolves around the two medium stage scenarios and the test cases examined to verify their completion.

Scenario 3 – Notifications and alerts for (at-risk) individuals

Scenario 1 includes the steps that have to be performed in order to onboard an individual and a carer to the system, which results in the acquisition of personal data that are used to classify (at-risk) individuals into specific profiles (“€œpersonas”). The accumulated data is then used (in an anonymous manner) by the Care Service Providers (CSPs), in conjunction with external data, for further analysis and for issuing simple notifications. In parallel, cares are registered to the platform and are connected to the (at-risk) individuals.

Regarding the test cases, first, the framework for an outlier detection model was created, delineating sets of persona outlier identification medical rules per condition or per group of conditions or of combination of conditions facilitating the definition of “personas outliers”. Subsequently the development of the outlier algorithmic and notification model was evaluated. The SHAL Web App received data from the defined sources and executed on Aegis the Outlier Notification Model in a configurable period of time. The model identified at-risk individuals / outliers by processing the new data against the defined rules. Additionally, template notifications that relate to medical rules have been created and have been linked to the different personas.


Scenario 4 – Smart home comfort profiling and notifications

Scenario 4 pertains the comfort profile model fitting framework and associated notifications to the end users, and constitutes the intermediate level of the Smart Home service of the demonstrator. The profiling process concentrates on the identification of personalized preferences for an at-risk person, based on the monitored environmental conditions and operational status, when he/she is at his/her living premises. Such comfort profiles are subsequently enriched by limits regarding VOC conditions, extracted from respective standards and directives. Given these models, the real-time data are continuously examined, and notifications/alerts, shown to the mobile app of the individual and/or informal carer, are generated, when conditions are recognized as not comfortable or potential detrimental to the person’™s health. The respective user stories are the following:

  • CSP: In order to enable the offering of the envisioned service, the CSP needs to estimate the personal preferences of individuals with respect to indoor living conditions. To that extent, the data scientist, working for the CSP, develops a comfort-profiling algorithm which is periodically trained on the smart home data, after the processing steps described in Scenario 2. The profiling framework is implemented in the AEGIS platform, allowing thus scalability of the used algorithmic processes. The model is then employed to continuously predict discomfort/potentially unhealthy indoor conditions, from the incoming data streams, and alert either the individual or the carer, through the mobile app, about the identified risks.
  • At-risk individual/Carer: The person and/or his/her informal carer register to the SHAL notification service, which allows them to receive the generated alerts prescribed above.

Two test cases were defined for evaluation of scenario 4. Firstly, The comfort profiling algorithm was created as a Jupyter notebook in the AEGIS platform. The required data-streams were previously established and data were cleaned and normalized. Consequently, the algorithm estimates the required Bayesian classifier model parameters and exports the class conditional posterior probabilities for given temperatures. Through these, the algorithm identified the comfort limits, which are utilized in the following test case to generate warnings on adverse indoor conditions. Following the identification of the comfort limits, a notification process was established within the smart home gateway, which checks in real-time the indoor environmental conditions and posts warning notifications to the SHAL backbone server, upon noticing that these deviate outside the acceptable limits.


The test cases were completed with success. At the final stage of the evaluation process, an internal focus group, comprised of six participants, was organized among the data scientists and developers of the demonstrator, so as to perform a further qualitative analysis. The details can be found in the associated AEGIS Deliverable 5.4.