Identification of respiratory effort sleep araousal using machine learning

Sleep is an indispensable physiological process for human health and quality of life. A good night's sleep directly affects not only an individual's physical well-being but also their mental and emotional balance. The early diagnosis and proper analysis of sleep disorders are of great importance in improving an individual's quality of life. One of the most common methods for detecting sleep disorders is the analysis of polysomnography (PSG) recordings obtained by monitoring an individual throughout the night. These multidimensional data, obtained from various sensors such as EEG, EMG, respiration, heart rate, and body position, are typically analyzed manually, which is a time-consuming and error-prone process. In this study, the aim was to detect sleep arousal events in real-time using PSG recordings and to automate this process using machine learning algorithms. The dataset was obtained from 113 individuals at the Yozgat Bozok University Chest Diseases Sleep Laboratory using the Philips Alice Clinic PSG device. The data collection process was carried out with the approval of the relevant ethics committee, and the recordings were manually annotated by a specialist physician according to the American Academy of Sleep Medicine (AASM) criteria. A total of 513.5 million sensor data points were analyzed using various machine learning algorithms, and classification performance was evaluated using data processing techniques. The Extra Tree Classification method achieved a success rate of 96.81%. The results demonstrate that machine learning techniques integrated with PSG data offer an effective approach for reliably and quickly detecting sleep disturbances. This study provides an important foundation for automated systems that could contribute to the early diagnosis of sleep disorders.