COVID-19 Detection from Cough Sounds Using XGBoost and LSTM Networks
Elmoundher Hadjaidji, Mohamed Cherif Amara Korba, Khaled Khelil · Traitement du signal · 2024
COVID-19, a contagious respiratory virus with symptoms like a dry cough, prompted intensive diagnostic efforts.Current standards fall short in controlling transmission, driving researchers to explore automated identification methods.In this work, using artificial intelligence and audio signal processing techniques, an automated system is developed.After extracting cough segments from audio recordings through the eXtreme Gradient Boost algorithm, the system attempts to detect COVID-19 employing a deep learning-based approach known as Long Short-Term Memory.In particular, the XGBoost model identifies the cough segment, and the LSTM-based model conduct binary classification on it to establish whether a person is positive or negative for COVID-19.To assess the proposed detection scheme, several experiments were conducted with the use of two publicly available cough sound datasets, namely COUGHVID and VIRUFY, which were collected from coronavirus-infected and non-infected persons through a large-scale crowdsourced campaign.The suggested system results were validated through comparisons with prior studies, demonstrating its strong performance even in noisy environments.Additionally, the obtained results indicate that the proposed method for detecting COVID-19 performs admirably under ideal conditions, achieving approximately 97% accuracy on the VIRUFY dataset and an impressive classification rate of nearly 88% on the COUGHVID dataset.