Developing Smart and Unobtrusive Mobile Home Care: A Deep Learning Approach

Abstract

Chronic conditions, frailty, dementia, and other diseases or symptoms significantly affect senior citizens’ health, safety, and quality of life. The insufficient labor supply of the home care market requests the involvement of modern information technology such as sensors, Internet of Things, and artificial intelligence. Healthcare providers and information systems (IS) researchers have sought to develop mobile home care approaches that adopt sensing technology to improve their effectiveness and efficiency. However, existing economics and behavioral methodologies for Health Information Technology (HIT) and health data analytics approaches were not designed for mobile data. Novel computational Information Technology (IT) artifacts are required to address significant mobile home care needs. Given the societal importance of mobile home care, this dissertation presents four essays that follow design science guidelines to design IT artifacts for mobile home care applications. Essay I employs a Sequence-to-Sequence model to extract high-level ADLs (e.g., cooking) from multi-modal smart home sensors (e.g., motion, on/off sensors). Essay II develops a hierarchical ADL recognition framework to generate interpretable intermediate results as well as a novel interaction kernel for Convolutional Neural Networks (CNNs) to extract the interaction semantics. Essay III leverages deep transfer learning to address object motion sensor data scarcity and improve human identification performance for customized home care. Essay IV designs an attention mechanism-based framework to better interpret deep learning-based models for mobile home care. Beyond the practical contributions provided to home care practitioners, this dissertation offers numerous design principles to guide future mobile health analytics IS research.