Sleep-Wake Classification Using Acceleration Data from Wearable Wrist Worn Sensors: REVIEW

Abstract

AbstractSleep study has been the most important field of research in recent times. Various works have been published to date that tries to detect the sleep/wake stage differences as well as various stages of sleep. Polysomnography (PSG) is a clinical procedure to score sleep by trained technicians. However, it is not very comfortable for the subject due to its set up and the subject must be placed away from their normal sleeping environment. The identification of wake or sleep stages at home requires the use of the multimodal wearable sensors like ActiGraph, Apple watch, etc. From these devices data is collected for categorization of sleep-wake states as it is easy as well as comfortable method and the subject can use it without going away from his/ her natural sleeping environment. This method also does not require the supervision from trained personnel. These wearable sensors provide the parameters like the acceleration data, heart rate, skin conductance, etc., from which features are extracted that acts as an input used by the different classifiers for detection. Deep learning algorithms like Long Short Term Memory neural networks have been found as a novel application in this classification problem of sleep/wake stages. The validation of various sleep detection systems is either validated against the ground truths like the clinical polysomnography, sleep diaries or can be validated against a self-proposed evaluation metrics. This paper gives a brief overview of the state-of-the-art works that have been done to date in sleep classification.KeywordsWearable sensorsAccelerometer dataDeep learning methodSleep/wake classification