130
D. Zekri et al.
The smart home concept includes homes equipped with simple environmental sensors and more complex systems including audio, video and biometric systems. The raw information captured by the sensors can obviously not be shared
as such with the medical staff or used directly to detect changes in behavior
automatically. On the contrary, extracted knowledge could be used to enrich the
information displayed to the medical staff and improve the precision of early
detections. There is evidence that opportunistic home surveillance prevents in
some cases hospitalization.
In this paper, we focus on the problem of learning from smart home sensor
data describing elderly’s activities. Our objective in this work is to propose an
approach to identify periods of time when behavior changes occur and detect
anomalies in this period (e.g., the elderly sleep less and less every month). Our
contributions in this paper can be summarized as follows.
1. We model a behavior pattern using training dataset, defined as the user’s
usual activities in his/her daily routine.
2. We calculate a daily score by comparing activity patterns. This daily score
variation provides a global vision of the behavior of the elderly person over a
period of time.
3. We detect anomalies related to every activity in the period of behavioral
deviation.
The rest of this paper is organized as follows. In Sect. 2, we discuss related
works. In Sect. 3 we present our approach. In Sect. 4 we report the experimentation of our proposal on real datasets. In Sect. 5 we present our conclusions and
some research directions.
2 Related Works
With the use of smart homes, the daily activities and behavioral patterns of
residents can now be monitored through sensors embedded within various areas
in the home. This allows elderly people to be more independent while providing assistance to their family and caregivers. In this section, we describe some
research works regarding the analysis of behavior and health monitoring for
elderly people in the smart home context.
Works in [1] use anomaly detections on wearable sensors to provide an intelligent living environment for elderly residents. The detection of anomalies is
based on several parameters: location, time, duration, type of activity and transitions between activities. The experiments provided consist in a semi-supervised
learning approach.
[2] and [3] study elderly residents diagnosed with dementia living independently in real home environments. They applied respectively neural networks
and clustering algorithms to predict sensor activity. When an error is detected,
timely audio or visual prompts are sent to the dementia patients.
Gjoreski et al. [4] have proposed a system to monitor users’daily activity by
combining accelerometers with an electrocardiogram (ECG) sensor. Measured
D. Zekri et al.
The smart home concept includes homes equipped with simple environmental sensors and more complex systems including audio, video and biometric systems. The raw information captured by the sensors can obviously not be shared
as such with the medical staff or used directly to detect changes in behavior
automatically. On the contrary, extracted knowledge could be used to enrich the
information displayed to the medical staff and improve the precision of early
detections. There is evidence that opportunistic home surveillance prevents in
some cases hospitalization.
In this paper, we focus on the problem of learning from smart home sensor
data describing elderly’s activities. Our objective in this work is to propose an
approach to identify periods of time when behavior changes occur and detect
anomalies in this period (e.g., the elderly sleep less and less every month). Our
contributions in this paper can be summarized as follows.
1. We model a behavior pattern using training dataset, defined as the user’s
usual activities in his/her daily routine.
2. We calculate a daily score by comparing activity patterns. This daily score
variation provides a global vision of the behavior of the elderly person over a
period of time.
3. We detect anomalies related to every activity in the period of behavioral
deviation.
The rest of this paper is organized as follows. In Sect. 2, we discuss related
works. In Sect. 3 we present our approach. In Sect. 4 we report the experimentation of our proposal on real datasets. In Sect. 5 we present our conclusions and
some research directions.
2 Related Works
With the use of smart homes, the daily activities and behavioral patterns of
residents can now be monitored through sensors embedded within various areas
in the home. This allows elderly people to be more independent while providing assistance to their family and caregivers. In this section, we describe some
research works regarding the analysis of behavior and health monitoring for
elderly people in the smart home context.
Works in [1] use anomaly detections on wearable sensors to provide an intelligent living environment for elderly residents. The detection of anomalies is
based on several parameters: location, time, duration, type of activity and transitions between activities. The experiments provided consist in a semi-supervised
learning approach.
[2] and [3] study elderly residents diagnosed with dementia living independently in real home environments. They applied respectively neural networks
and clustering algorithms to predict sensor activity. When an error is detected,
timely audio or visual prompts are sent to the dementia patients.
Gjoreski et al. [4] have proposed a system to monitor users’daily activity by
combining accelerometers with an electrocardiogram (ECG) sensor. Measured