Forecasting the behavior of an elderly using wireless sensors data in a smart home

Abstract

In this paper, the ability to determine the wellness of an elderly living alone in a smart home using a low-cost, robust, flexible and data driven intelligent system is presented. A framework integrating temporal and spatial contextual information for determining the wellness of an elderly has been modeled. A novel behavior detection process based on the observed sensor data in performing essential daily activities has been designed and developed. The developed prototype is used to forecast the behavior and wellness of the elderly by monitoring the daily usages of appliances in a smart home. Wellness models are tested at various elderly houses, and the experimental results are encouraging. The wellness models are updated based on the time series analysis.

Introduction

The age span of elderly people is increasing, and this trend set to continue in future (United Nations (UN), 2011). A normal elderly person is assumed to be well if he/she performs basic daily activities at regular intervals of time. This implies that the wellness of a person can be assessable and can be quantified in terms of wellness indexes. The elderly people prefer to live independently, but the self-regulating way of life involves with risks. In particular, the daily home activity involving basic functions like preparing food, showering, walking, sleeping, watching television, reading books etc., is a key indicator in determining the performance of home activities of an elderly and consequently their wellness. Thus an intelligent, data driven engineering system is needed to record the basic activities of elderly at home. The analysis of data in real-time will determine whether any change of regular activities have taken place and if any preventive action is required. This precautionary measure may help in reducing the future health-care cost.

A variety of sensing systems for monitoring and assessing functional abilities of elderly behavior in a smart home have been developed (DeSilva et al., 2012, Dan et al., 2011, Cook, 2012). Behavior prediction methods relating to abnormal behavior with temporal rules have been proposed (Tibor et al., 2011, Noury and Hadidi, 2012). Nouri and Hadidi (2012) have “demonstrated the feasibility to produce simulated data which mimics the data gathered by presence of sensors in field conditions”; and “imagined to raise an alarm whenever the real collected data becomes significantly different from the simulated data”. Tibor et al. (2011), have introduced a generic ambient agent-based model to study the dynamic patterns of human. Simulation experiments have been conducted with the generic ambient agent-based model and the outcomes have been formally analyzed. However, these methods will lead to a high number of false alarms when their behavior prediction techniques do not satisfy the conditions of the knowledge base. It is useful, to be able to determine an event on certain conditions using predicting techniques. Thus, Predictive Ambient Intelligence (PAI) techniques are used in a smart home environment in order to forecast the behavior of inhabitant under a monitoring environment.

A Predictive Ambient Intelligence environment gathers information from Wireless Sensor Networks (WSN) including environmental changes and occupants' interactions with the objects within the monitoring environment. Collected data are used to determine the behavior of inhabitant at different times by using prediction methods. The prediction involves the extraction of patterns related to sensor activations. This is then used to classify the sequence of activities and match it to predict the next activity (Das and Cook, 2005). Healthcare specialists believe that the best procedures to recognize health conditions of elderly before they become sick is to look for the changes in the actions of everyday life such as activities of daily living (ADLs), Instrumental ADLs(IADLs) (Tapia et al., 2004, Lawton and Broody, 1969, James, 2008, Rogers et al., 1998). Effective classification of ADLs is an important factor for detecting the changes in the routine habits of the elderly to determine their health conditions. There have been a momentous research activities in activity recognition and anomaly detection subdomains of a smart home monitoring system (Tapia et al., 2004, Dan et al., 2011, Cook, 2012, Hu and Yang, 2008, Zhongna et al., 2009, Marie et al., 2012).

The general structure of a Smart Home Monitoring System (SHMS) involves smart devices, a communication system and an intelligent system. In general, “Smart Home” is an expression utilized for dwellings outfitted with technologies that enable proper scrutiny of residents promoting autonomy and upholding of better health. A wide range of smart home research is carried out in the world and is reported in (e.g., Tapia et al., 2004, Dan et al., 2011, Cook, 2012, Hu and Yang, 2008; Zhongna et al., 2009; Marie et al., 2012). The focus of researches is quite different as every researcher has distinct requirements with varied research objectives. However, from the government point of view they would like to cut cost on healthcare by using technology in healthcare.

One of the main objectives towards the development of a smart home monitoring system is to have a minimum number of sensors and perform intelligent data analysis (Okour et al., 2012). “Networks cannot be deployed massively, if they are not affordable”. Moreover, healthcare providers will not use them in a clinical routine, if they are not cost-effective or do not have important benefits over the traditional health care delivery (Romero et al., 2008). A home monitoring system based on a minimum number of sensors will lead to the development of a low-cost system. A low-cost system will be affordable to the elderly who are mostly retired or possibly a low-income earner. Though a significant amount of research has been done already and is continuing, there is still a need for an effective technique and a low-cost solution for elderly care in smart home monitoring. The problem is not simple as human behavior is quite complex and is not possible to define by computer algorithms guided by associated rules. The patterns of sensor values must be able to provide information related to the identification of elderly activities to determine the wellbeing of elderly at a macro-level.

The wellness monitoring system for an elderly is confined to monitoring the performance of daily activities and decides whether the behavior is regular or irregular. A key element in the success of a well-being monitoring systems depends on understanding of the normal lifestyle and the degree to which the behavior of the elderly has deviated from that norm. There are several research works related to ambient monitoring systems like CASALA (Brain et al., 2011), BT Care (Nick et al., 2005), and other related works using different approaches such as validation concepts for short and long term ambient health monitoring (Chiriac et al., 2012); and analyzing the behavioral patterns through health status monitoring (Barger, 2005). In this paper, we propose the big picture of measuring the overall wellness level of an elderly and also the micro scale level of behavior recognition, activity annotation and so on. The research findings discussed in this paper can be deemed to provide a solution to a question raised in the paper (Dan et al., 2011) on real-time data analysis of sensor technology for smart homes. Dan et al. (2011) have expressed their concern on the technical issue to be addressed before sensor technology can be successfully deployed into real-world residential settings. The application of the proposed prototype will provide sufficient evidence for detecting changes in daily routines associated with functional activities using sensor data.

The most difficult task in an ambient intelligence environment is the prediction of behavioral patterns from the sensory observations. The accumulated data from sensory devices in a Wireless Sensor Network can become huge and complex if many sensor devices are used in a smart home. It is extremely difficult to find the correct number of sensors required for a smart home monitoring. The reported literature suggests a large number of sensors for monitoring the elderly behavior; subsequently the systems are very expensive and are unattractive to the elderly (Okour et al., 2012). The sensors should be invisible and have wider acceptance to the elderly (Alemdar and Ersoy, 2010). The current research targets the transformation of an existing-home to a smart home using the WSN based home monitoring system. The sensors data analysis to extract knowledge and adapt to changes in elderly behavior can provide a solution to the wellness determination of the elderly problem.

The remainder of this paper is organized as follows: Section 2 discusses some related works. Section 3 describes the developed system and the implementation details. Section 4 provides the results of verification of the methods on a publicly available data set and consequently testing the system at different elderly houses. Finally, Section 5 provides conclusions.