國立高雄大學圖資館 |

Language: English

Back

Comparative Analysis of Machine Learning Models for Sleep Apnea and Hypopnea Detection Using Pulse Oximetry and Heart Rate Variability /

Record Type:	Electronic resources : Monograph/item
Title/Author:	Comparative Analysis of Machine Learning Models for Sleep Apnea and Hypopnea Detection Using Pulse Oximetry and Heart Rate Variability /Vedang Sharma.
Author:	Sharma, Vedang,
Description:	1 electronic resource (78 pages)
Notes:	Source: Masters Abstracts International, Volume: 87-06.
Notes:	Advisors: Shandhi, Md. Mobashir Hasan; Bansal, Ajay Committee members: Alhindawi, Nouh.
Contained By:	Masters Abstracts International87-06.
Subject:	Biomedical engineering.
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=32285268
ISBN:	9798265458728

Comparative Analysis of Machine Learning Models for Sleep Apnea and Hypopnea Detection Using Pulse Oximetry and Heart Rate Variability /
Sharma, Vedang,

Comparative Analysis of Machine Learning Models for Sleep Apnea and Hypopnea Detection Using Pulse Oximetry and Heart Rate Variability /Vedang Sharma. - 1 electronic resource (78 pages)

Source: Masters Abstracts International, Volume: 87-06.

Obstructive Sleep Apnea and Hypopnea Syndrome (OSAHS) is a prevalent and significantly underdiagnosed global health issue. The condition carries risks for dangerous and sometimes life-threatening complications, such as heart failure, stroke, and sudden cardiac death. Traditional diagnosis relies on in-laboratory polysomnography (PSG), a method that is costly, resource-intensive, and not suitable for large-scale public screening. This thesis investigates the efficacy of machine learning (ML) models in developing a scalable, accessible, and reliable screening tool for OSAHS.This study conducts a rigorous comparative analysis of multiple ML algorithms, including Random Forest, Support Vector Machine, and eXtreme Gradient Boosting (XGBoost). The models are trained and validated using non-invasive physiological signals, specifically pulse oximetry (SpO2) and heart rate variability (HRV). This analysis is derived from a large, well-characterized public dataset of 6,399 overnight patient recordings.Models were developed using a robust nested cross-validation approach on a dedicated training set, which comprised 80% of the data. The models were then independently validated on a held-out test set of the remaining 20% to ensure generalizability. The XGBoost model achieved the best performance, yielding a high F1-score of 0.897 on the independent test set.Furthermore, this work systematically evaluates model robustness under simulated real-world constraints. These constraints include reduced data availability, with recording lengths from eight hours down to 30 minutes, and degraded signal fidelity, with SpO2 sampling rates down to 0.25 Hz. Model explainability, explored using SHapley Additive exPlanations (SHAP), confirmed that predictions were driven by clinically relevant markers, particularly SpO2 entropy, desaturation events, and median saturation levels.The findings demonstrate the strong potential of using machine learning with signals obtainable from consumer-grade wearable devices. This approach can pave the way for an effective, scalable screening paradigm for OSAHS, enabling earlier detection and treatment.

English

ISBN: 9798265458728Subjects--Topical Terms:

190330
Biomedical engineering.
Subjects--Index Terms:

Machine learning

Comparative Analysis of Machine Learning Models for Sleep Apnea and Hypopnea Detection Using Pulse Oximetry and Heart Rate Variability /
LDR:03645nmm a22004213i 4500 001 693782
005 20260511063024.5
006 m o d
007 cr|nu||||||||
008 260819s2025 miu||||||m |||||||eng d
020 $a 9798265458728
035 $a (MiAaPQD)AAI32285268
035 $a AAI32285268
040 $a MiAaPQD $b eng $c MiAaPQD $e rda
100 1 $a Sharma, Vedang, $e author. $3 1010459
245 1 0 $a Comparative Analysis of Machine Learning Models for Sleep Apnea and Hypopnea Detection Using Pulse Oximetry and Heart Rate Variability / $c Vedang Sharma.
264 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2025
300 $a 1 electronic resource (78 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Masters Abstracts International, Volume: 87-06.
500 $a Advisors: Shandhi, Md. Mobashir Hasan; Bansal, Ajay Committee members: Alhindawi, Nouh.
502 $b M.S. $c Arizona State University $d 2025.
520 $a Obstructive Sleep Apnea and Hypopnea Syndrome (OSAHS) is a prevalent and significantly underdiagnosed global health issue. The condition carries risks for dangerous and sometimes life-threatening complications, such as heart failure, stroke, and sudden cardiac death. Traditional diagnosis relies on in-laboratory polysomnography (PSG), a method that is costly, resource-intensive, and not suitable for large-scale public screening. This thesis investigates the efficacy of machine learning (ML) models in developing a scalable, accessible, and reliable screening tool for OSAHS.This study conducts a rigorous comparative analysis of multiple ML algorithms, including Random Forest, Support Vector Machine, and eXtreme Gradient Boosting (XGBoost). The models are trained and validated using non-invasive physiological signals, specifically pulse oximetry (SpO2) and heart rate variability (HRV). This analysis is derived from a large, well-characterized public dataset of 6,399 overnight patient recordings.Models were developed using a robust nested cross-validation approach on a dedicated training set, which comprised 80% of the data. The models were then independently validated on a held-out test set of the remaining 20% to ensure generalizability. The XGBoost model achieved the best performance, yielding a high F1-score of 0.897 on the independent test set.Furthermore, this work systematically evaluates model robustness under simulated real-world constraints. These constraints include reduced data availability, with recording lengths from eight hours down to 30 minutes, and degraded signal fidelity, with SpO2 sampling rates down to 0.25 Hz. Model explainability, explored using SHapley Additive exPlanations (SHAP), confirmed that predictions were driven by clinically relevant markers, particularly SpO2 entropy, desaturation events, and median saturation levels.The findings demonstrate the strong potential of using machine learning with signals obtainable from consumer-grade wearable devices. This approach can pave the way for an effective, scalable screening paradigm for OSAHS, enabling earlier detection and treatment.
546 $a English
590 $a School code: 0010
650 4 $a Biomedical engineering. $3 190330
650 4 $a Bioinformatics. $3 194415
653 $a Machine learning
653 $a Cardiac death
653 $a Heart rate variability
690 $a 0800
690 $a 0541
690 $a 0715
710 2 $a Arizona State University. $b Software Engineering. $e degree granting institution. $3 1010460
720 1 $a Shandhi, Md. Mobashir Hasan $e degree supervisor.
720 1 $a Bansal, Ajay $e degree supervisor.
773 0 $t Masters Abstracts International $g 87-06.
790 $a 0010
791 $a M.S.
792 $a 2025
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=32285268