Impact of Data Pre-Processing on Covid-19 Diagnosis Using Machine Learning Algorithms

Abstract

Abstract: Human coronaviruses present a significant disease burden. Identifying infected coronavirus patients using artificial intelligence draws researchers’ attention all over the world. Blood test is a striking element that can significantly contribute to provide a reliable, accurate, and quick automated detection tool of covid-19 diagnosis. Medical datasets are known to be associated with different data problems mainly, unbalancing, missing values, and amplitude variations. Performance of classifiers cannot be correctly assessed without handling those problems. For this, the paper at hand proposed multiple solutions that merge several data pre-processing techniques with three dominant classifiers namely Deep Learning (DL), K-Nearest Neighbors (KNN), and Support Vector Machines (SVM). After detailed dataset treatment, all three classifiers achieved good performance according to the gold standard with SVM scoring the highest accuracy and sensitivity of 86% and 95% respectively. This study showed the clinical soundness and feasibility of utilizing blood test analysis and machine learning as a replacement to rRT-PCR for detecting COVID-19-positive cases.<br>