Real-time sleep posture classification using wearable accelerometers and machine learning models

Abstract

Sleep posture plays a critical role in sleep quality and health, influencing conditions such as sleep apnea. Accurate classification of sleep postures is essential for diagnosing and treating sleep-related disorders. The sleep posture can be detected by using wearable acceleromter. This paper presents an realtime classification system for four sleep postures by integrating accelerometer data with a machine learning (ML) model. The proposed system was tested with various ML models, including decision trees (DT), random forest (RF), K-nearest neighbors (KNN), support vector classifier (SVC), and logistic regression (LR), across multiple performance metrics. The results demonstrate that the LR model, when combined with accelerometer data, significantly outperforms other methods, achieving a classification accuracy of 91%. This paper also discusses the system’s potential for real-time deployment on embedded devices, contributing to advancements in sleep posture monitoring.