Processing and classification of COVID-19 audio data with optimization-based learning techniques

COVID-19, which emerged in late 2019, has entered the literature as a global disease. COVID-19 is a rapidly spreading and difficult to control disease. For this reason, millions of people lost their lives due to this virus during those times. The virus appears with respiratory tract disease symptoms and is transmitted through contact. Today, this virus has been brought under control and prevented with vaccines. In this study, negative and positive labeled datasets containing COVID-19 breathing and coughing sounds were used. Two separate approaches were put forward with two separate datasets. The sounds in the datasets were preprocessed and meaningful features were obtained. Then, approaches were applied with deep learning and machine learning methods. In the first approach, feature extraction was performed before the audio data. 12 different feature extractions were used and these were combined. Adam optimization accompanied the CNN and MLP classifiers for 1224 features. CNN and MLP were optimized with a learning rate of 0.0001. Approximately 84% accuracy was achieved for one-dimensional CNN and approximately 92% for MLP. In the second approach, a dataset was obtained by combining two different datasets. 5 different feature extractions were made to this dataset. In total, 66 features were obtained for each sound. Then, a classification result was obtained by applying only RF. Then, the best parameter values were determined with cPSO and RF was applied. Feature selection was also made with bPSO. The best and most meaningful features were extracted for 66 features. 44 features were extracted as a result of bPSO. RF was applied with the features obtained with bPSO. High performance was obtained as a result of three modeling. 94% accuracy was obtained for only RF, 95% for RF-cPOS, and 96% for RF-bPSO. When all results were compared, it was seen that RF-bPSO provided the best classification from the features obtained vectorially.