Analysis of COVID-19 Coughs: From the Mildest to the Most Severe Form, a Realistic Classification Using Deep Learning

Cough is the most recurrent symptom of lung disease. In addition, COVID-19 is an unparalleled lung disease. The spread of this pandemic has resulted in over 600 million positive cases and over 6 million deaths worldwide. Therefore, an efficient, inexpensive, and ubiquitous diagnostic tool is essential to help fight lung disease and the COVID-19 crisis. Deep learning and machine learning algorithms can be used to analyze the cough sounds of infected patients and infer predictions. We made use of constructivist logic. The cough data are from our research lab and the COUGHVID research lab. This Diagnostic approach, based on deep learning and feature extraction from Mel spectrograms, can recognize cough sounds from sick people without COVID-19, with severe and mild COVID-19, and also recognize cough sounds from healthy patients. The model used is a variant of ConvNet. The quiet environments, which allowed the acquisition of data, reduce systematic and random errors in the quality of the audio. The architecture of the convolutional neural networks exploited, gives an overall Accuracy of 90.33%. This system could have a significant positive social impact by minimizing transmission of the virus, speeding up patient treatment, and freeing up hospital resources. In addition, early diagnosis of COVID-19 may also prevent further disease progression and improve the effectiveness of screening efforts.