A home daily activity simulation model for the evaluation of lifestyle monitoring systems
Introduction
In England, over the next fifty years, the number of people over 65 is expected to rise by 56% and a similar trend is observed in most other western countries. As a consequence, researchers and governments are looking at novel solutions to support people in their own home by automatically monitoring their daily activities [2], [3], [6], [9], [12], [15], [18], [20]. As part of the newer generations of telecare, lifestyle monitoring (LM) aims to observe the activities of older or vulnerable individuals and if circumstances change determine if a medical or care intervention may be beneficial. Generally, LM uses a set of sensors fitted in the house and aims to detect those deviations from ‘normal’ behaviour that could be indicative of a change in care needs (e.g. Mobility problems, difficulty of toileting, etc.).
LM has attracted a lot of interest in the last few years, however, most of the research publications in this domain present results based on only a small number of users. Indeed, a review of the literature suggests that to the end of 2009 only 4 trials have been conducted with more than 20 participants [5]. The limited scale and scope of trials can largely be attributed to the difficulty of performing such experiments. Indeed, while it is desirable to evaluate a lifestyle monitoring system under real conditions, several issues generally arise which act to limit the scope of the trials. These include:
- a.
Difficulty in recruiting participants who will accept a relatively intrusive system installation in their home without direct immediate return.
- b.
Data should be observed over a long period of time.
- c.
Difficulties in collecting ground-truth information. Indeed, to validate and develop a system that is supposed to observe individual activity, it is essential to know which activity that individual is actually involved in at any time. This information could be collected by means of diaries, but these are not always accurate and can be very demanding for the participants if they need to be maintained over a long period of time. Another approach could be to visually monitor and manually annotate individual activity using video cameras on site, but this can be considered intrusive and would require a laborious video transcription and annotation phase.
- d.
In order to validate a system that aims to detect abnormality, a significant number of abnormal events must be observed. By definition, such abnormal events are rare.
Given these constraints, it becomes clear that trials with users, while ultimately essential, can also be limited, especially at the early stages of development. As a consequence, thorough and large scale testing on synthetic data can be of great benefit to the development of novel LM systems.
By using a simulation tool, it becomes possible to simulate virtually any condition, and it becomes possible to test the effect of any change on the system or on subject behaviour. With a simulator it is possible to create a change in behaviour on demand, rather than running a real system while hoping to encounter specific types of change. Likewise, a simulator could provide information on the effect of including a new sensor (with known specifications) in the system before having to encounter the costs in both time and money of real world experiments.
Verone et al. [19] attempted to simulate behavioural data from a patient living in an intelligent home. Their paper presents some interesting ideas and appears to provide a useful background in developing the simulator being proposed here. However, Verone et al. only simulate room transitions based on daily behaviour profiles. More importantly, they do not propose any mechanism to modify behaviour according to specific changes in the subject′s condition or care need. More recently, Noury et al. [13] proposed a similar approach based on Markov models corresponding to activities corresponding to seven periods of the day. This model do not allow long term dependences of activities to be taken into account, i.e. the decision on the activity to perform at time t is only dependant on the activity performed at time t−1. In this paper, a simulator is proposed that can serve to generate synthetic data of daily activity which could be used as a tool for the evaluation of LM systems. The evaluation of such a system should ideally involve trials with users in real situations. Issues regarding the evaluation of medical informatics are discussed in [10]. As argued by the authors in [1], computer simulation provides a flexible approach to evaluation in health informatics.
The proposed simulation model aims to be used to evaluate the capability of LM systems to detect unusual change in activities indicative of changes in health conditions. To achieve this, we aim to generate realistic sequences of daily activity. The simulator allows simulating circumstances such as a reduction in mobility or illness that could lead to a change in care need. With such a scheme it becomes possible to evaluate abnormality detection algorithms but also the effect of a new sensor, before physically integrating them, and consequently could significantly reduce development costs.
It is believed that the modelling of the daily pattern of activities is the key step toward the development of the simulator and can be considered as the most challenging aspect. Indeed, the activities undertaken by an individual during a day are driven by a number of factors and, unlike machines, human behaviour can be unpredictable. In terms of the simulation this unpredictability is reproduced through the use of stochastic models.
Section snippets
Method
The primary objective of the simulator is to generate data that can be used for evaluating the performance of a LM system and should be able to reproduce sensor activations that correspond to specific user behaviours. Note that there is a differentiation between those key features which are characteristic of behaviour or the system and those key parameters that translate these features into the values used by the simulator. Key features could for example be the number of times the subject has a
Results
Simulation model verification is often defined by reference to Sargent [16] as “ensuring that the computer program of the computerised model and its implementation are correct”. A formal definition of model validation is given by Schlesinger et al. [17]: “substantiation that a computerized model within its domain of applicability possesses a satisfactory range of accuracy consistent with the intended application of the model”. For the daily activity simulator, there is no set of specific tests
Discussion and conclusion
The proposed model appears to be satisfactory for the intended application however it can be considered to suffer from number of limitations compared to a more realistic simulation of daily activities that could be required in other applications. The model is built to reproduce realistic statistical features, however, it does not integrate firm limitations encountered in real life. For example, it might be impossible to go shopping after 8 pm because the shop is closed. To cope with this
Conflict of interest statement
None declared.
References (21)
Evaluation in health informatics: computer simulation
Comput. Biol. Med.
(2002)Evaluation: salvation or nemesis of medical informatics?
Comput. Biol. Med.
(2002)- et al.
Computer simulation of the activity of the elderly person living independently in a Health Smart Home
Comput. Methods Programs Biomed.
(2012) - et al.
Human activity recognition in pervasive health-care: supporting efficient remote collaboration
J. Netw. Comput. Appl.
(2008) - et al.
Lifestyle monitoring technology for supported independence
Comput. Control Eng. J.
(1998) - et al.
Health-status monitoring through analysis of behavioral patterns
IEEE Trans. Syst. Man Cybern.
(2005) - et al.
An evaluation of second and third generation telecare services in older people′s housing
J. Telemed. Telecare
(2008) - et al.
A systematic review of lifestyle monitoring technologies
J. Telemed. Telecare.
(2013) - et al.
Modelling of behavioural patterns for abnormality detection in the context of lifestyle reassurance, in Progress in Pattern Recognition
(2008) - et al.
Introduction to Algorithms
(2001)
Cited by (9)
Adaptive monitoring system for e-health smart homes
2018, Pervasive and Mobile ComputingCitation Excerpt :It is worth mentioning that the use of simulated data in health informatics research is common practice [26]. It has emerged as an alternative approach for developing and validating health monitoring systems [27,28 ]. Simulation methods can handle different activity trends thus avoiding tedious lab experiments or real-world deployments.
Smart Indoor Space Simulation Methodologies: A Review
2022, IEEE Sensors JournalAn activity rule based approach to simulate ADL sequences
2018, IEEE AccessSmart homes and sensors for surveillance and preventive education at home: Example of obesity
2016, Information (Switzerland)Persim 3D: Context-Driven Simulation and Modeling of Human Activities in Smart Spaces
2015, IEEE Transactions on Automation Science and EngineeringAn indoor-movement simulator for ambient assisted living systems
2015, 2015 IEEE Globecom Workshops, GC Wkshps 2015 - Proceedings
- 1
Now at: Sony, European Technology Center, 70327 Stuttgart, Germany.