Optimal human sleep awakening with alarm adjusted by real-time tracking and machine learning

Abstract

75% of youth snooze their wake-up alarms. Some more than others. There exist consumer electronics that can sense light sleep, and initiate a wake-up sequence prematurely. Is it, however, possible to predict whether or not someone will snooze? Predicting an outcome is not necessarily constrained to magic tricks if machine learning can be applied to your problem. That's what is investigated in this thesis: The feasibility of predicting if a person will snooze, and for how long, by learning from his/her sleep data. Data is recorded by a bed-side device that can be operated by anyone. The goal is to improve the user's sleep by waking them up in a light sleep stage, no matter if this happens right before, or right after the alarm is set to. The ultimate motivation is improved health. Snoozing is a problem that deserves more attention, this paper deploys the best of sleep science, machine learning, and embedded systems to propose a possible solution.