New Pathways for Unsupervised Machine Learning in Digital Health : Applications and Future Potentials for Sleep

Abstract

Unsupervised machine learning has emerged as a powerful method, offering the ability to detect patterns and reveal hidden relationships in health-related data without relying on manual labels. This thesis aims to explore the existing applications and future potential of unsupervised machine learning in digital health. There are vast amounts of unlabeled data in medical research, and there is a growing need for digital solutions to tackle health challenges in a growing and aging population. Digital health plays a key role in the paradigm shift of medicine towards predictive, preventive, personalized, and participatory healthcare. By investigating unsupervised machine learning in sleep research, this thesis aims to derive implications for digital health and contribute to the fields of information systems and data science. Sleep, as one of the pillars of health, has a substantial contribution to various mechanisms in our body, including the brain, hormonal balance, and the cardiovascular system. Good sleep can help to improve overall physical and mental health, while poor sleep is associated with chronic diseases, decreased cognitive function, and a shortened lifespan. Even though it is common knowledge that sleep is important, it can be difficult to maintain good sleep. There can be physical, psychological and lifestyle factors impacting the quality of sleep. Digital health and machine learning can address this issue from various perspectives, such as the efficient diagnosis and provision of treatment options for people with sleep disorders, the collection of longitudinal sleep data, and the analysis of sleep recordings. This thesis first maps all existing publications on unsupervised machine learning in sleep research and then conducts four case studies with selected unsupervised machine learning methods on different forms of sleep data. The cases use anomaly detection, dimensionality reduction, clustering, and association rules with data on respiration and brain during sleep, as well as objective and subjective sleep quality assessment, and engagement with a digital therapeutics application. Each case aims to represent a novel method to show the range and diversity of contributions of unsupervised machine learning for sleep and encourage researchers to tread new pathways for digital health. The main contributions of the thesis are outlining the existing applications of unsupervised machine learning in sleep research, exploring and applying different forms of health data, critically discussing the clinical value of unsupervised learning, and, ultimately, finding new pathways for unsupervised learning in digital health.