Abstract

A comprehensive understanding of the risk factors related to coronavirus disease 2019 (COVID-19), which can cause various severe clinical pictures, is vital for predicting the progression of the disease and administering the appropriate treatment to prevent adverse outcomes. Therefore, this study aimed to predict the presence of COVID-19 and to identify possible risk factors for COVID-19 patients using the CatBoost machine learning (ML) technique. CatBoost was used to classify COVID-19. The model results were evaluated using sensitivity (Sen), specificity (Spe), negative predictive value (NPV), positive predictive value (PPV), accuracy (Acc), F1score, precision, recall, and area under the curve (AUC) metrics. In the modeling phase, a10 - fold cross-validation approach was employed. Lastly, variable importance values were derived through modeling. The dataset used in the article is open access and was collected in India in May 2020 this data is available at https://www.kaggle.com/code/anirbansarkar823/covid-presence-detection-ensembling. When the results of The CatBoost model successfully classified the modeling dataset, with Acc, Sen, Spe, PPV, NPV, F1score, recall, precision, and AUC values of 97.79%, 97.21%, 98.55%, 93.23%, 98.77%, 98.66%, 96.00%, 97.79%, and 99.62%, respectively. Considering the findings show that the risk factors related to COVID-19 infection, which has become a threat to humanity, can be successfully determined. Similar studies can update information on the transmission routes of the disease and positively change its course.

Key words: Diarrhea, machine learning, random forest, bagged CART, CatBoost model