An efficient algorithm to measure quality and quantity of sleep using smartphone

Abstract

Sleep quality and quantity a ects an individual's personal health. Polysomnography (PSG) is the conventional approach for sleep monitoring. PSG has several limitations in terms of sleep monitoring by a regular user. A number of wearable devices with embedded sensors have emerged in recent past as an alternative to polysomnography (PSG) for regular sleep monitoring directly by the user. These devices are intrusive and cause user discomfort besides being expensive. We present an algorithm to detect sleep using a smartphone with the help of its inbuilt accelerometer sensor. We exploit three di erent approaches to classify data into two states, Sleep and Wake. The approaches vary in terms of their adaptability. One is based on Kushida equation and uses a xed threshold. Second is based on statistical method and also uses a xed threshold but the threshold is decided dynamically at the time of data processing. Finally the third is based on Hidden Markov Model (HMM) training. The rst one being least adaptable, the second one is moderate, and the third one is the most adaptive. The complexity also increases in the same order. Dataset consists of sleep data for four subjects for twelve days each collected using our android application for data collection. The accuracy of sleep detection of each of the three approaches is then compared with that of Zeo sensor, which is a medically approved device to measure sleep. We nd that HMM training classi cation approach is as much as 84% accurate. It is more accurate as compared to Kushida equation based on xed threshold and statistical method based on xed threshold. In order to concisely represent the sleep quality of a person, we model the sleep data of a user using HMM. The parameters obtained from HMM can further be used to concisely represent sleep quality of an individual in the form of a model speci c to that individual. We present an analysis to nd out tradeo between the amount of training data and the accuracy provided in modeling the sleep. We nd that six days of sleep data is su cient for accurate modeling. We compare accuracy of our HMM training algorithm with a representative third party application SleepTime available from Google Play Store for Android. We nd that the classi cation done using HMM approach is closer to that done by Zeo by 13% as compared to the third party Android application SleepTime. In order to get an estimate as to how our classi cation algorithm will perform on a smartphone we compare the performance of our algorithm on two di erent PC architectures one with more and the other with less resources. Results suggest that the algorithm is feasible enough for execution on an android smartphone.