Mobile Health Analytics for Senior Care: A Data Mining and Deep Learning Approach

Abstract

Senior citizens confront numerous challenges to their independent living, including chronic physical health conditions and a decline in mobility. With the advancement of mobile sensing technologies, medical professionals and information systems (IS) researchers have sought to apply data mining techniques to provide precise, prompt, and personalized assessment for falls and health conditions including Parkinson’s disease. Given the societal importance of senior care, my dissertation aims to address the following four research questions: (1) how can we promptly detect senior citizens’ adverse events, e.g., falls, to alleviate consequences, (2) how can we precisely assess senior citizens’ health risks, e.g., fall risks, to provide proper interventions, (3) how can we leverage multiple data sources and assess senior citizens’ health risks in a more holistic manner, and (4) how can we profile senior citizens’ long-term health progression for more personalized care. This dissertation presents four essays to tackle these questions. The essays develop state-of-the-art data mining and deep learning techniques to address selected senior care inquiries. The first essay focuses on a novel hidden Markov model with sensor orientation calibration to detect falls. The second essay presents a two-dimensional heterogeneous convolutional neural network to precisely assess fall risks. The third essay leverages deep multisource multitask learning to achieve sensor fusion and assess multiple health risks and disease severities. The final essay develops an adaptive time-aware convolutional long short term memory model that enables long-term health profiling with time irregularities. Presented frameworks, systems, and design principles not only advance mobile health analytics and deep learning methodologies, but also guide future computational design science research in IS.