Study of inferring sleep stages using wearable sensors in home setting

Abstract

Sleep plays a vital role in ones mental as well as physical health. It helps the brain work properly. In this research, we seek to find the optimal placement of sensors for predicting sleep stages and verification of these stages are performed with the help of home polysomnogram. Compared to sleep labs which may cause discomfort to participants, due to new place and multiple sensors connected to them, the experiment data is collected at home using noninvasive sensors. Sleep stages are calculated by the electrical activity in brain using electroencephalogram (EEG) signal. Through this study, we provide quantitative and qualitative analysis for feature selection and analyze why sensor placement on some parts of body is better than other. Using machine learning algorithms, we predicted the accuracy of Sleep and Wake stages being 85.97%, accuracy of Wake, REM and NREM stages being 63.94% and 48.3% accuracy for all sleep stages including N1, N2 and N3 using IMU and PPG sensors without the help of brain waves (EEG signals).