JMIR Publications

ABSTRACT

Background:

Our research collects approximately 2 years of data from real smart home environments of independently living elderly for monitoring activities of daily living activities (ADL) and develops a method that detects deviations from identified living routines, referred to as anomalies in daily living behaviors.

Objective:

The research objectives are to: • Collect data from sensors and devices in multiple smart homes environment, • Identify behavioral patterns in ADL of elderly, • Detect anomalies in ADLs and • Send notifications in real-time to elderly participants’ family members. We propose a framework for identifying patterns and anomalies in activities of daily living from real-world sensor data collected from sensors installed in 12 apartments, in northern Sweden, in uncontrolled settings over approximately two years. We use a statistical approach to detect anomalies based on the time duration of each activity. Notifications about normal daily activities and anomalies are sent via SMS to the elderly relatives as both positive and negative notifications via our developed real-time online system, respectively.

Methods:

We extract features from sensor data by calculating the time durations and frequency of visits. We used a parametric statistical approach to detect anomalies based on unusually short or long durations by estimating parameters such as mean (μ), and standard deviation (σ) over hourly time windows varying from 80 to 360 days for different households. The confidence level is at least 75% of the tested values lies within two (σ) within the mean. If the actual duration was outside the limits of two standard deviations [μ − 2σ, μ + 2σ] or if the activity didn’t occur, the absence of an activity, then an anomaly was triggered.

Results:

The patterns detected from sensor data matched the self-reported routines of the users. Our system generated approximately 1000 notifications which were sent to 9 households within 45-65 days A service evaluation regarding the received notifications showed a positive user experience, we received an average 4 on a 1-5 scale. More than 75% of the notifications of each household fall within two standard deviations of the mean implying they correspond to their normal behavior.

Conclusions:

In this research we developed, implemented and evaluated a real-time monitoring system with real elderly participants in an uncontrolled environment. We used different off-the-shelf sensors and IoT devices installed in elderly homes. We developed an SMS-based notification service and conducted user evaluations. This service acts as an extension to existing health/social care services operated by the municipality of Skellefteå, thus helping the elderly to live in their homes independently and support relatives of the elderly in caregiving activities.