Ambient Sleep Quality Analysis with a Machine Learning Model

The average person moves about 40 to 50 times a night during sleep. In addition to complete body movements, twitches and limb jerks are common during normal sleep. Chest or abdomen also move slightly and periodically due to respiration. Passive infrared sensors can detect fine motion such as respiration induced chest motion as well as stronger motion induced by body movements. This study presents a machine learning model for sleep quality assessment using ambient passive infrared sensors. The ambient sensor and associated machine learning algorithm successfully identified sleep and motion activity with better than 88% accuracy. The model presented here performed better in monitoring motion than a wearable Garmin watch and reported respiration rates inline with a medical quality wearable sleep monitoring sensor by Masimo Inc.