Supporting Healthy Aging with Active Assisted Living Technology: From Developing Smart Home Data Ecosystem to Activities of Daily Living Recognition

Abstract

Background: The aging population is growing rapidly worldwide, posing significant challenges for healthcare systems. This demographic shift demands innovative approaches to address the unique physical, cognitive, and social needs of older adults.. Traditional healthcare systems are often ill-equipped to provide continuous monitoring and support, leading to increased pressure on caregivers and healthcare providers. Active Assisted Living technologies, combined with Artificial Intelligence, present an opportunity to bridge this gap by integrating advanced sensors and analytics into everyday living environments. These technologies can enhance healthcare monitoring, promote independent living, and improve overall well-being for aging populations. The aim of this thesis is to develop and evaluate AI-enhanced AAL systems that address key challenges in aging healthcare. This includes constructing robust data ecosystems for sensor integration, advancing methods for physical indicator prediction, Activities of Daily Living recognition, and ensuring the adaptability of these technologies for real-world applications.

Methods: This thesis employed a multi-phase research methodology, integrating systematic reviews, infrastructure development, and experimental evaluations to advance Active Assisted Living technologies for aging populations. The research began with a comprehensive scoping review to systematically map the landscape of AI applications in AAL technologies. The review followed the PRISMA methodology, with searches conducted across six databases: PubMed, CINAHL, IEEE Xplore, Scopus, ACM Digital Library, and Web of Science. Studies published over the past decade were considered, resulting in 64 eligible papers that were reviewed and analyzed to identify critical gaps. Building on these findings, a cloud-native AAL data ecosystem was developed to facilitate the collection, integration, and management of sensor data from commercially available AAL devices. This ecosystem was structured across four layers: perception, transportation, processing, and application. Leveraging cloud computing technologies, the system ensured scalability, interoperability, and real-time data processing. Advanced data acquisition and management protocols were implemented to address challenges related to reliability, security, and compliance with health data standards. The functionality of the data ecosystem was validated through a heart rate prediction study using contactless monitoring techniques. Forty participants performed 23 predefined daily living activities in a smart home environment equipped with Swidget ambient sensors and Empatica E4 wristbands. Data collected from ambient sensors were analyzed using machine learning models, such as Random Forests, to predict heart rates accurately. Further research investigated Activities of Daily Living recognition using the same data assets. A proof-of-concept study involving younger adults in controlled smart home environments evaluated the effectiveness of contactless ambient sensors. Machine learning models, such as Adaptive Boosting, Decision Trees, and Gaussian Naive Bayes, were employed to classify daily activities, achieving high accuracy. To address the challenge of generalizability for Activities of Daily Living recognition among population, a hybrid transfer learning framework was developed. This framework combined Correlation Alignment and Conditional Adversarial Domain Adaptation techniques to adapt models trained on younger adults for application in older populations. Experiments involving 26 healthy older adults were conducted in semi-controlled environments, focusing on less structured activities to simulate real-world conditions.

Results: The scoping review identified critical gaps in the application of AI within AAL technologies, particularly the lack of centralized data ecosystem and real-world experimentation. The developed cloud-native AAL data ecosystem successfully integrated data from multiple commercially available sensors, facilitating seamless data processing and real-time analytics. The heart rate prediction study achieved a high prediction performance with a Mean Absolute Error of 6.023 using the Random Forest model. The ADLs recognition study, conducted with younger adults in controlled settings, demonstrated an overall accuracy of 0.964 using the Adaptive Boosting model, significantly outperforming wearable sensor-based methods and hybrid modes. The hybrid transfer learning framework further addressed the generalizability challenge by adapting models trained on younger adults in controlled settings to older adults in semi-controlled settings. The hybrid adaptive model achieved an accuracy of 0.576 and an F1-score of 0.538, compared to baseline methods, which achieved only 0.098 in accuracy the most. Additionally, incorporating a small amount of target population data in training enhanced the performance, with accuracy increasing by a maximum of 0.344.

Conclusion: This thesis makes significant contributions to the advancement of AAL technologies on public health monitoring by addressing foundational gaps in data ecosystem development, health indicators prediction, ADLs recognition, and its transferability to older adults in real-world. The research demonstrates the feasibility of AI-enhanced AAL systems to provide reliable, scalable, and inclusive solutions for healthcare monitoring and activity recognition in aging populations. By developing and validating methodologies for real-world application, this work also highlights the potential of these technologies to support public health initiatives. The integration of AAL systems into public health frameworks could enable proactive monitoring, early detection of health risks, and personalized interventions, ultimately improving population-level health outcomes. This thesis lays the groundwork for empowering older adults to live independently, safely, and with improved quality of life, while also supporting caregivers and healthcare providers with actionable insights.