AdvSleepNet: Accelerometer-Based Sleep Stage Detection

In this paper, a neural network model named AdvSleepNet is proposed for the accurate identification of various sleep stages, including wake, rapid eye movement, and non-rapid eye movement stages. This research pioneers a noninvasive approach utilizing wrist-wearable accelerometers, addressing limitations associated with traditional methods. Leveraging the potential of this technology, the aim is to provide valuable information about sleep quality and enable early detection of sleep-related issues by developing a sleep stage detection model using accelerometer data from a wrist-worn device. This research culminates in the AdvSleepNet neural network model, which demonstrates exceptional performance compared to all applied classifiers. Models trained on Apple Watch data are evaluated using the independent dataset of the Multi-ethnic Study of Atherosclerosis to extend the findings. Similar accuracy in predicting sleep is achieved when compared with testing on Apple Watch data. These results emphasize the viability and efficacy of utilizing wearable technology for comprehensive sleep analysis, potentially revolutionizing the field of sleep surveillance while providing valuable insights into individuals’ sleep patterns and overall health.