Machine learning-based classifiers for obstructive sleep apnea diagnosis using ECG and EDR signals

Simple item page
dc.contributor.advisorTobón Llano, Luis Eduardo
dc.contributor.advisorJaramillo Otoya, Laura
dc.contributor.authorMoreno Granja, Johny
dc.contributor.authorOcaciones García, Alexis
dc.date.accessioned2026-09-03T15:25:48Z
dc.date.available2026-09-03T15:25:48Z
dc.date.issued2026
dc.description.abstractObstructive sleep apnea (OSA) is one of the most common sleep-related breathing disorders with serious cardiovascular and neurological implications. It is normally diagnosed in the context of polysomnography (PSG) which is very expensive, cumbersome, and invasive in the process. In this work, we applied machine learning and deep learning techniques to develop an automated ECG method for detecting and assessing the severity of OSA to make diagnosis more accessible and efficient. We used the PhysioNet Apnea-ECG database and used 10 classification methods: logistic regression, support vector machine with RBF kernel, random forest, CNN and RNN with raw ECG, CNN and RNN models with RR intervals and ECG-derived respiration (RR+EDR) and three ensemble strategies: soft voting, AUC-weighted voting, and stacking. Heart rate variability and EDR features were extracted from one-minute ECG windows with a ±1-minute temporal context. All the methods have subject-independent data partitioning to avoid data leakage and to generate more realistic models. The models were evaluated internally and independently using the official x01–x35 test set and 35 subjects and 12,600 one-minute windows. Performance was assessed based on precision, sensitivity, specificity, accuracy, F1 score, AUC, and Cohen's kappa coefficient. The patient-level hierarchical analysis was also conducted, where the minute-level predictions were collected into an Apnea Index and four categories of severity were established: normal, mild, moderate and severe. The results showed that classical machine learning and ensemble approaches provided the most consistent generalization. The Weighted-AUC set had the best overall external performance with an AUC of 0.939 and F1 score of 0.834. Patient severity estimation also showed promising results with a mean absolute error of the apnea index of 3.01 events per hour and a severity accuracy of around 80% in the external assessment.eng
dc.description.degreelevelMaestría
dc.description.degreenameMagíster en Ingeniería
dc.format.extent103 p.
dc.format.mimetypeapplication/pdf
dc.identifier.urihttp://hdl.handle.net/11522/5780
dc.language.isospa
dc.publisherPontificia Universidad Javeriana Cali
dc.publisher.departmentFacultad de Ingeniería y Ciencias
dc.publisher.programMaestría en Ingeniería
dc.rights.accessrightshttp://purl.org/coar/access_right/c_abf2
dc.rights.creativecommonshttps://creativecommons.org/licenses/by-nc-sa/4.0/
dc.rights.urihttps://creativecommons.org/licenses/by-nc-sa/4.0/
dc.subjectApnea obstructiva del sueñospa
dc.subjectECGspa
dc.subjectEDRspa
dc.subjectClasificadoresspa
dc.subjectAprendizaje automáticospa
dc.subjectAprendizaje profundospa
dc.subjectObstructive sleep apneaeng
dc.subjectClassifierseng
dc.subjectMachine learningeng
dc.subjectDeep learningeng
dc.titleMachine learning-based classifiers for obstructive sleep apnea diagnosis using ECG and EDR signalseng
dc.typemaster thesis
dc.type.coarhttp://purl.org/coar/resource_type/c_bdcc
dc.type.localTesis/Trabajo de grado - Monografía - Maestría
dc.type.redcolhttps://purl.org/redcol/resource_type/TM
Files
Original bundle
Now showing 1 - 2 of 2
No Thumbnail Available
Name:
Trabajo de grado en pdf - MACHINE LEARNING-BASED CLASSIFIERS.pdf
Size:
1.88 MB
Format:
Adobe Portable Document Format
Download
No Thumbnail Available
Name:
Licencia de uso - MACHINE LEARNING-BASED CLASSIFIERS FOR OBS.pdf
Size:
282.65 KB
Format:
Adobe Portable Document Format
Download
License bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
license.txt
Size:
1.71 KB
Format:
Item-specific license agreed to upon submission
Description:
Download
Collections
Maestría en Ingeniería