Feasibility of motion sensor-based human activity recognition for supporting independence in smart homes

Objectives

This study investigates the feasibility of leveraging Internet of Things (IoT) and artificial intelligence (AI) technologies to accurately identify key human activities critical for evaluating the ability of older adults to live independently.

Methods

Seventeen clinically relevant activities for assessing independent living of older adults were identified using standard clinical tools and insights from in-depth discussions with clinicians in geriatric care. Real-world data were then collected within a home environment using 20 IoT wearable and object sensors, as healthy adult participants performed each activity. A comprehensive set of features was extracted from the collected data, using both time and frequency domains, which were then used to develop a random forest machine learning algorithm to accurately recognize these activities.

Results

Data from 10 participants (7 male, 3 female) showed that the proposed algorithm achieved an average accuracy of 87.5 % in recognizing 17 key activities. Additionally, the algorithm achieved 97.95 % accuracy in identifying four functional areas namely mobility, hygiene, nutrition and hydration, and medication intake.

Discussion

These findings highlight the potential of IoT and AI-driven technologies to identify key activities essential for independent living in a home environment. The high accuracy of activity recognition enables reliable monitoring, offering valuable insights into the capabilities of older adults to live independently. This approach could be used to empower caregivers and clinicians to deliver timely, tailored care, fostering improved support and quality of life.