Clinical diagnosis of overlapping symptoms in COVID-19 based on machine learning model

Abstract

The global pandemic COVID-19 erupted and infected an estimated 10% of the worlds population. Since vaccinations greatly reduced hospitalization rates, most countries removed the restrictive policies implemented to combat the virus. It has become a rather common illness with more than twelve thousand active hospitalizations. As a result, convenient COVID-19 diagnosis from diseases that display overlapping symptoms has become increasingly important. An effective method for patient self-diagnosis greatly reduces hospital presentation, saving time and medical resources. This study uses machine learning techniques to classify and predict several common respiratory diseases quickly and accurately. The author trains several machine learning models that attempt to predict four diseases based on their distinct clinical signs. An open-access database on Kaggle developed for this disease classification is selected and further processed via principal component analysis to decrease database dimension and pinpoint critical symptoms. Support Vector Machine Classifier (SVM), Nave Bayes (NB), Logistic Regression (LR), and Random Forest (RF) models are used, and their performances are compared. Study results show that the LR model slightly outperforms the others. In conclusion, the effectiveness of the proposed method is proved for classifying the symptoms of patients with allergies, colds, flu, and Covid-19 in this study.