Home // HEALTHINFO 2024, The Ninth International Conference on Informatics and Assistive Technologies for Health-Care, Medical Support and Wellbeing // View article

Real-world ADL Recognition with Deep Learning and Smartwatches: A Pilot Study

Authors:
Mustafa Elhadi Ahmed
Hongnian Yu
Michael Vassallo
Pelagia Koufaki

Keywords: Activities of Daily Living; Activity Recognition; Deep Learning; Independent living; Smartwatch.

Abstract:
The global aging population poses significant challenges to healthcare systems, especially in promoting independent living and reducing caregiver burdens. Technology-Enabled Care (TEC), which leverages digital tools and Artificial Intelligence (AI), has emerged as a promising solution to support older adults. A crucial component within TEC is the automatic recognition of Activities of Daily Living (ADLs), essential for early detection of health declines and personalized care. Traditional ADL recognition research, often conducted in controlled environments, does not adequately address real-world complexities. This study bridges the gap between laboratory prototypes and practical applications by developing a user-friendly ADL recognition framework using commercial smartwatches. A hybrid model, combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks, was trained on accelerometer and gyroscope data to recognize activities like dishwashing and walking. Initially validated in a lab setting with an accuracy of 94%, the model was subsequently tested over a 20-day pilot study involving five participants (mean age = 32 years, SD = 4.5), each wearing an Apple Watch device. Real-world results revealed a significant performance drop, with accuracy declining to 81%. Activities like mopping maintained high recognition accuracy, while subtler tasks, such as walking and washing face posed challenges due to movement variability. These findings underscore the need for model optimization using real-world data to improve recognition accuracy and address variability in movement patterns. Further research is essential to refine these systems for broader applications, develop strategies to enhance user adherence, and ultimately support the independence and well-being of aging individuals.

Pages: 1 to 7

Copyright: Copyright (c) IARIA, 2024

Publication date: November 3, 2024

Published in: conference

ISSN: 2519-8491

ISBN: 978-1-68558-204-3

Location: Nice, France

Dates: from November 3, 2024 to November 7, 2024