Abstract:

Objective: The whole world has been trying to deal with the SARS-CoV-2 virus since December 2019. Early diagnosis is of great importance for physicians, as the early symptoms of the disease overlap with other common conditions such as cold and flu. It is aimed to develop a mobile application to predict the results of routine blood tests and SARS-Cov-2 (positive/negative) using machine learning algorithms with anonymous hospital data that is open to common sharing in this study. Materials and Methods: After eliminating the missing observation, class imbalance, outlier observation, and unrelated variable problems in the data set, the classification performances of machine learning methods were tested, and then a logistic regression model was established for the detection of COVID-19 with appropriate variables. Using this model, a machine learning-based mobile application has been designed. Results: The variables that gave the best results in diagnosis were eosinophils, leukocytes, thrombocytes, monocytes, red blood cells, and basophils. After solving the data pre-processing problems, the classification performance of the algorithms used has increased considerably compared to the performance values in the raw data. Conclusion: With the developed mobile application, it is possible to estimate the diagnosis of Covid-19 quickly and easily by using routine blood test results.

Keywords: