Investigation of Covid-19 Infection with Clinical Data Using Decision Trees

Abstract

The coronavirus disease, namely Covid-19 infection, which was declared a worldwide epidemic by the World Health Organization (WHO) in 2020, was first seen in Wuhan, China in the last months of 2019 and has affected the whole world. Early diagnosis of this rapidly spreading epidemic is important to prevent the disease. For this reason, methods such as image processing, deep learning, and machine learning have become important to detect the epidemic early. In this study, it has been tried to classify individuals who test positive and negative for Covid-19 based on some laboratory test results with several Decision Tree methods. Since the original form of the data set has an uneven distribution, the data set has been balanced by applying the oversampling and undersampling methods used for such data sets as a pre-processing study. Balanced dataset and original dataset using 5-Fold Cross Validation (CV), 10-Fold Cross Validation and Leave-One-Out (LOO)-CV, Random Forest (RF), Random Tree (RT), J48, ıt was analyzed with alternating decision tree (ADTree) and Function Trees (FT) classifiers. As a result of the examination, the most successful result was shown by the RF classifier with 87.5% success rates using CV-5 in the original data set, 93.3% using CV-10 and LOO-CV in the oversampling method, and 79% using CV-5 in the undersampling method. In addition to success rates, sensitivity-specificity metrics, which are important for patient and healthy diagnosis, were examined in terms of each classification algorithm and CV value.