Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
The role of features in predictive deep learning models for auditory tuberculosis classification
Thesis (MEng)--Stellenbosch University, 2024.
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English |
| Published: |
Stellenbosch : Stellenbosch University
2025
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867613815254810624 |
|---|---|
| access_status_str | Open Access |
| author | Knight, Michael Sean |
| author2 | Wolhuter, Riaan
|
| author_browse | Knight, Michael Sean Wolhuter, Riaan |
| author_facet | Wolhuter, Riaan
Knight, Michael Sean |
| author_sort | Knight, Michael Sean |
| collection | Thesis |
| dc_rights_str_mv | Stellenbosch University |
| description | Thesis (MEng)--Stellenbosch University, 2024. |
| format | Thesis |
| id | oai:scholar.sun.ac.za:10019.1/131717 |
| institution | Stellenbosch University (South Africa) |
| language | English |
| last_indexed | 2026-06-10T12:42:07.859Z |
| license_str | Other — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from SUNScholar — Stellenbosch University Repository |
| publishDate | 2025 |
| publishDateRange | 2025 |
| publishDateSort | 2025 |
| publisher | Stellenbosch : Stellenbosch University |
| publisherStr | Stellenbosch : Stellenbosch University |
| record_format | dspace |
| source_str | SUNScholar — Stellenbosch University Repository |
| spelling | oai:scholar.sun.ac.za:10019.1/131717 The role of features in predictive deep learning models for auditory tuberculosis classification Knight, Michael Sean Wolhuter, Riaan Niesler, Thomas Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering. Deep learning (Machine learning) Tuberculosis -- Diagnosis Medical informatics Predictive analytics UCTD Thesis (MEng)--Stellenbosch University, 2024. ENGLISH ABSTRACT: The application of machine learning to cough audio for the purpose of tuberculosis (TB) detection promises to be a low-cost and easily deployable diagnostic method. Some, but little, work has been performed on this topic and while it has shown potential, a lack of consistency and scope leave a lack of clarity regarding where new research should focus. This work starts by considering previously identified model architectures and features, in conjunction with a common dataset, and methodically analyses and compares the effectiveness of several approaches. Three model architectures: logistic regression (LR), ResNets and BiLSTMs are tested, and a modification of the ResNet architecture, termed SkipNet, is proposed. Three feature types: linear filter banks (LFBs), mel filter banks (MFBs) and mel frequency cepstral coefficients (MFCCs) are applied to the architectures which themselves are combined in three different ways: ensemble models, single models and teacher-student models. Finally, forward sequential search (FSS) is explored as a means of reducing computation and improving robustness to irrelevant input. From our experiments, we conclude that ResNets provide the best-performing classifier architecture achieving an AUC of 77.48%. Among features, LFBs and MFBs usually outperformed MFCCs, with MFBs normally performing slightly better overall. The ensemble and the teacher-student configurations achieved comparable performance, typically both outperforming the single model configuration comfortably, with the teacher-student configuration generally achieving the highest overall performance. Finally, FSS successfully enhanced the performance of key models, increasing the AUC of BiLSTMs from 72.91% to 77.09% and the AUC of ResNet-18s from 74.17% to 77.49%, while reducing the input dimensionality by at least 50% and, therefore, reducing computation. AFRIKAANSE OPSOMMING: Geen opsomming beskikbaar. Masters 2025-02-20T08:26:31Z 2025-02-20T08:26:31Z 2024-12 Thesis https://scholar.sun.ac.za/handle/10019.1/131717 en Stellenbosch University xiv, 79 pages : illustrations application/pdf Stellenbosch : Stellenbosch University |
| spellingShingle | Deep learning (Machine learning) Tuberculosis -- Diagnosis Medical informatics Predictive analytics UCTD Knight, Michael Sean The role of features in predictive deep learning models for auditory tuberculosis classification |
| title | The role of features in predictive deep learning models for auditory tuberculosis classification |
| title_full | The role of features in predictive deep learning models for auditory tuberculosis classification |
| title_fullStr | The role of features in predictive deep learning models for auditory tuberculosis classification |
| title_full_unstemmed | The role of features in predictive deep learning models for auditory tuberculosis classification |
| title_short | The role of features in predictive deep learning models for auditory tuberculosis classification |
| title_sort | role of features in predictive deep learning models for auditory tuberculosis classification |
| topic | Deep learning (Machine learning) Tuberculosis -- Diagnosis Medical informatics Predictive analytics UCTD |
| url | https://scholar.sun.ac.za/handle/10019.1/131717 |
| work_keys_str_mv | AT knightmichaelsean theroleoffeaturesinpredictivedeeplearningmodelsforauditorytuberculosisclassification AT knightmichaelsean roleoffeaturesinpredictivedeeplearningmodelsforauditorytuberculosisclassification |