Multi-aggregation Strategies in Ensemble-Based Machine Learning and Deep Learning Models for Cough-Based COVID-19 Detection
Lombarkia, Takieddine Marir, Naila Marir, Farid
Lombarkia, Takieddine
Marir, Naila
Marir, Farid
Type
Supervisor
Subject
Date
2025-04-21
Collections
Research Projects
Organizational Units
Journal Issue
Abstract
Respiratory diseases have become a major area of research, especially during the fight against the Coronavirus Disease (COVID-19) pandemic. While various measures have been taken to prevent and control the spread of the virus, the existing diagnostic methods can be time-consuming and invasive. This paper describes a novel framework utilizing Multi-Aggregation Strategies in Ensemble-Based Machine Learning and Deep Learning (MASE-MDL) to detect COVID-19 through cough analysis. Our method employs a comprehensive feature extraction process from cough audio signals. Initially, raw cough audio data undergo preprocessing steps, including noise reduction and signal normalization, to enhance signal quality and consistency. Subsequently, relevant features are extracted from the preprocessed audio signals. These features encompass a range of acoustic characteristics including, but not limited to, frequency spectrum, temporal patterns, and spectral entropy while capturing diverse aspects of cough sounds indicative of respiratory conditions. Moreover, to leverage the collective intelligence of multiple predictive models, such as machine learning algorithms and recurrent neural networks like GRU and LSTM, we employ a multi-aggregation strategy. This approach combines predictions from diverse models, each trained on distinct feature subsets or utilizing different learning algorithms. Tests on the CoughVid dataset have shown high accuracy, sensitivity, and specificity, indicating its potential as a reliable screening tool for healthcare professionals. Furthermore, this approach helps develop new diagnostic techniques that do not require laboratory tests or CT scans. Ultimately, this could lead to better health outcomes for patients and improved respiratory disease management in hospitals.