Bastani Najafabadi, Vahid2026-09-022026-09-022026-08-252026-08-252026-08-31http://hdl.handle.net/1993/40052Obstructive sleep apnea (OSA) is a common sleep disorder that remains underdiagnosed due to the cost, complexity, and limited accessibility of standard diagnostic methods such as polysomnography (PSG). Accessible and objective screening approaches to identify individuals in need of treatment for OSA before confirmatory sleep assessment will significantly help with the backlog of PSG sleep sessions. Previous studies have demonstrated the feasibility of OSA screening during wakefulness using tracheal breathing sounds (TBS); however, these frameworks were predominantly developed using specific recording hardware and acquisition configurations. Therefore, it remains unclear whether diagnostically relevant wakefulness TBS information can be preserved using a substantially lower-cost microphone with different electroacoustic characteristics and whether previously developed frameworks can be transferred directly to such recordings. Given that the presentation of OSA is affected by individual characteristics such as age, sex, body mass index (BMI), neck circumference, smoking history and craniofacial structure, screening frameworks should consider both physiological features and anthropometric covariates. This thesis develops a machine learning framework for wakefulness OSA screening using tracheal breathing sounds (TBS) recorded with a low-cost microphone in a practical clinical environment. Data of 247 individuals with various degrees of OSA severity were analyzed. Recorded data were segmented into inspiration and expiration phases, followed by acoustic features extraction, feature reduction, and classification. A two-level ensemble architecture was implemented. Nine sub-classifiers were stratified by anthropometric profiles, which each sub-classifier was constructed as an ensemble of bagged decision trees, with a final prediction via probability-based voting. The proposed algorithm achieved an accuracy of 77.1%, sensitivity of 84.3%, and specificity of 59.9%. In contrast, applying the previously developed AWakeOSA framework directly to the current recordings resulted in substantially lower performance, indicating limited transferability to the present acquisition setting. These findings demonstrate that diagnostically useful wakefulness TBS information can be obtained using a low-cost microphone while suggesting that changes in recording hardware and acquisition conditions may require adaptation of previously developed classification frameworks. Future studies should validate the framework using larger and more balanced independent datasets, investigate intermediate AHI cases, and directly evaluate microphone-related effects using paired recordings under controlled acquisition conditions.engObstructive Sleep Apnea (OSA)Wakefulness ScreeningTracheal Breathing SoundsWavelet Packet DecompositionMachine LearningProbability-Based Voting