Abstract
Sleep apnoea (SA) is a prevalent but underdiagnosed sleep-related breathing disorder associated with significant
cognitive, cardiovascular, and metabolic consequences. Polysomnography (PSG), the clinical gold standard for diagnosis, is
expensive, resource-intensive, and unsuitable for large-scale or continuous home-based monitoring. This paper presents a
real-time sleep apnoea detection framework based solely on oxygen saturation (SpO₂) signals, enabling a simplified and noninvasive diagnostic approach. Experiments were conducted using recordings from the publicly available CAP Sleep
Database hosted on PhysioNet, which contains annotated overnight polysomnography data acquired in controlled clinical
environments. SpO₂ signals were segmented into standard 30-second windows, and time-domain statistical features were
extracted to characterise oxygen desaturation patterns. A Random Forest classifier was employed to discriminate apnoearelated desaturation events from normal breathing episodes.