Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Robust COVID-19 Detection from Cough Sounds using Deep Neural Decision Tree and Forest: A Comprehensive Cross-Datasets Evaluation

Rofiqul Islam, Nihad Karim Chowdhury, Muhammad Ashad Kabir

TL;DR

This work tackles COVID-19 detection from cough sounds using a robust DNDT and DNDF framework. It integrates RFECV feature selection, Bayesian optimization for hyper-parameters, SMOTE balancing, and ROC-AUC-based threshold moving, validated across five large cough datasets (Cambridge, Coswara, COUGHVID, Virufy, NoCoCoDa) and a combined dataset. The study demonstrates strong performance (AUC up to 0.99 on some datasets) and highlights cross-dataset transfer challenges, showing that combining datasets improves generalizability. The results indicate a promising, non-invasive screening approach with potential for broad real-world impact, while also underscoring the importance of dataset diversity to mitigate biases. The methodology advances cough-based COVID-19 diagnostics by delivering a rigorous, cross-datasets evaluation and a robust, ensemble-based classifier architecture.

Abstract

This research presents a robust approach to classifying COVID-19 cough sounds using cutting-edge machine-learning techniques. Leveraging deep neural decision trees and deep neural decision forests, our methodology demonstrates consistent performance across diverse cough sound datasets. We begin with a comprehensive extraction of features to capture a wide range of audio features from individuals, whether COVID-19 positive or negative. To determine the most important features, we use recursive feature elimination along with cross-validation. Bayesian optimization fine-tunes hyper-parameters of deep neural decision tree and deep neural decision forest models. Additionally, we integrate the SMOTE during training to ensure a balanced representation of positive and negative data. Model performance refinement is achieved through threshold optimization, maximizing the ROC-AUC score. Our approach undergoes a comprehensive evaluation in five datasets: Cambridge, Coswara, COUGHVID, Virufy, and the combined Virufy with the NoCoCoDa dataset. Consistently outperforming state-of-the-art methods, our proposed approach yields notable AUC scores of 0.97, 0.98, 0.92, 0.93, 0.99, and 0.99 across the respective datasets. Merging all datasets into a combined dataset, our method, using a deep neural decision forest classifier, achieves an AUC of 0.97. Also, our study includes a comprehensive cross-datasets analysis, revealing demographic and geographic differences in the cough sounds associated with COVID-19. These differences highlight the challenges in transferring learned features across diverse datasets and underscore the potential benefits of dataset integration, improving generalizability and enhancing COVID-19 detection from audio signals.

Robust COVID-19 Detection from Cough Sounds using Deep Neural Decision Tree and Forest: A Comprehensive Cross-Datasets Evaluation

TL;DR

This work tackles COVID-19 detection from cough sounds using a robust DNDT and DNDF framework. It integrates RFECV feature selection, Bayesian optimization for hyper-parameters, SMOTE balancing, and ROC-AUC-based threshold moving, validated across five large cough datasets (Cambridge, Coswara, COUGHVID, Virufy, NoCoCoDa) and a combined dataset. The study demonstrates strong performance (AUC up to 0.99 on some datasets) and highlights cross-dataset transfer challenges, showing that combining datasets improves generalizability. The results indicate a promising, non-invasive screening approach with potential for broad real-world impact, while also underscoring the importance of dataset diversity to mitigate biases. The methodology advances cough-based COVID-19 diagnostics by delivering a rigorous, cross-datasets evaluation and a robust, ensemble-based classifier architecture.

Abstract

This research presents a robust approach to classifying COVID-19 cough sounds using cutting-edge machine-learning techniques. Leveraging deep neural decision trees and deep neural decision forests, our methodology demonstrates consistent performance across diverse cough sound datasets. We begin with a comprehensive extraction of features to capture a wide range of audio features from individuals, whether COVID-19 positive or negative. To determine the most important features, we use recursive feature elimination along with cross-validation. Bayesian optimization fine-tunes hyper-parameters of deep neural decision tree and deep neural decision forest models. Additionally, we integrate the SMOTE during training to ensure a balanced representation of positive and negative data. Model performance refinement is achieved through threshold optimization, maximizing the ROC-AUC score. Our approach undergoes a comprehensive evaluation in five datasets: Cambridge, Coswara, COUGHVID, Virufy, and the combined Virufy with the NoCoCoDa dataset. Consistently outperforming state-of-the-art methods, our proposed approach yields notable AUC scores of 0.97, 0.98, 0.92, 0.93, 0.99, and 0.99 across the respective datasets. Merging all datasets into a combined dataset, our method, using a deep neural decision forest classifier, achieves an AUC of 0.97. Also, our study includes a comprehensive cross-datasets analysis, revealing demographic and geographic differences in the cough sounds associated with COVID-19. These differences highlight the challenges in transferring learned features across diverse datasets and underscore the potential benefits of dataset integration, improving generalizability and enhancing COVID-19 detection from audio signals.
Paper Structure (31 sections, 9 equations, 7 figures, 9 tables)

This paper contains 31 sections, 9 equations, 7 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Research questions
  4. Methodology
  5. Dataset description
  6. Cambridge dataset
  7. Coswara dataset
  8. COUGHVID dataset
  9. Virufy dataset
  10. NoCoCoDa dataset
  11. Combined dataset
  12. Feature extraction
  13. Mel-Frequency Cepstral Coefficients (MFCCs)
  14. Mel-Scaled Spectrogram
  15. Tonal centroid
  16. ...and 16 more sections

Figures (7)

  • Figure 1: An overview of the proposed method to classify COVID-19 based on the analysis of cough sound audio recordings.
  • Figure 2: Overview of the extraction process for 193 features across five types: MFCCs, Mel-Scaled Spectrogram, Tonal Centroid, Chromagram, and Spectral Contrast.
  • Figure 3: Feature selection with the recursive feature elimination with cross-validation (RFECV) and Extra-Trees classifier.
  • Figure 4: The confusion matrices for strategy 5, using the Deep Neural Decision Tree (DNDT) classifier with 10-fold cross-validation: (a) Cambridge asymptomatic, (b) Cambridge symptomatic, (c) Coswara, (d) COUGHVID, (e) Virufy, and (f) Virufy merged with NoCoCoDa. "C" represents COVID-19 Positive and "N" represents Non COVID-19 cough instances.
  • Figure 5: The confusion matrices for strategy 5, using the Deep Neural Decision Forest (DNDF) classifier with 10-fold cross-validation: (a) Cambridge asymptomatic, (b) Cambridge symptomatic, (c) Coswara, (d) COUGHVID, (e) Virufy, and (f) Virufy merged with NoCoCoDa datasets. "C" represents COVID-19 Positive, and "N" represents Non COVID-19 cough instances.
  • ...and 2 more figures