Abstract:

The COVID-19 pandemic has spread rapidly over the past two years, and the study has also increased. The distinction between patients with COVID-19 and non-sick is also one of the most important problems in the pandemic. If the early diagnosis of the disease is necessary, the automatic detection of COVID-19 is of great importance in terms of the risk of infection to non-patients. Diagnosis of the disease has made it much more difficult to see different symptoms and even develop without any symptoms. This has increased the work on the diagnosis of the disease. It is also possible to diagnose using the important features that exist in the respiratory records, such as cough sound recordings, one of these applications. In this study, automatic COVID-19 disease was detected using cough sound recordings. The COVID-19 Positive and Negative Patients' Cough Recordings (HIMANSHU) data set was used to first extract the Mel-Frequency Kepstrum Scorecards (MFCC) from audio files. The MFCC properties of different numbers are made equal by using statistical values. The spectral characteristics obtained by the MFCC method have been extracted in four different lengths to be 8, 16, 32, 64. Finally, the machine learning algorithms have been diagnosed using artificial nerve networks (YSA), Naive Bayes (NB), K-en Near Neighbor algorithms (kNN), and Rastgele Forest (RO) algorithms. In the study, two classes were used in the form of COVID or COVID-DEGIL. The application is operated with ten cross-verification methods. At the end of the study, the highest classification was achieved with the interpersonal kNN algorithm at 99.39%.

Keywords: