Machine Learning-Based Sleep Apnea Pre-Screening Using a Low-Cost Wearable in Thailand

Sleep apnea is a common sleep disorder that requires early identification to reduce associated health risks. Polysomnography (PSG), the clinical gold standard for diagnosis, is costly and has limited availability, particularly in resource-constrained healthcare settings. This study investigates an e-health-oriented approach for sleep apnea pre-screening by integrating data from a low-cost wearable oximetry device with machine learning models in a real clinical environment in Thailand. Data were collected from 192 participants at the Thammasat Hospital Sleep Laboratory using the Wellue O2 ring, which records oxygen saturation, pulse rate, and motion during sleep. Wearable-derived physiological signals, together with demographic and clinical information, were used as inputs to machine learning models, while PSG served as the reference standard. Support vector machine (SVM), random forest (RF), and one-dimensional convolutional neural network (1D-CNN) models were evaluated. The random forest model demonstrated the most favorable balance between accuracy and feasibility, highlighting the potential of wearable-assisted machine learning to support accessible e-health sleep apnea pre-screening.