COMPARATIVE ANALYSIS OF MACHINE LEARNING ALGORITHMS FOR CLASSIFYING CHEST X-RAY IMAGES OF COVID-19

Abstract

The fast spread of COVID-19 has made it essential to have quick and accurate ways to diagnose the disease using medical images. In this study, we present a comparative analysis of six machine learning algorithms for classifying chest X-ray images into three categories: COVID-19, pneumonia, and normal. The algorithms evaluated include Support Vector Machine (SVM), Random Forest, Decision Tree, K-Nearest Neighbors (KNN), Naive Bayes, and Logistic Regression. The models were trained and validated using a dataset of labeled X-ray images, with performance assessed based on key metrics such as validation accuracy, test accuracy, Matthews Correlation Coefficient (MCC), and ROC-AUC scores. Among the models, the SVM algorithm achieved the highest validation accuracy of 92.6% and test accuracy of 94.0%, with an MCC of 0.8785 and a macro-average ROC-AUC of 0.9877. Logistic Regression and Random Forest followed closely, with Logistic Regression attaining a test accuracy of 93.8% and Random Forest achieving 92.6%. K-Nearest Neighbors showed moderate performance, while Naive Bayes exhibited the lowest accuracy, likely due to its simplistic assumptions. Overall, the SVM model outperformed the others, demonstrating its robustness in distinguishing between the three categories in chest X-ray images. This comparative study highlights the potential of machine learning techniques in automating the detection of COVID-19 and pneumonia from radiographs, contributing to faster and more accurate diagnostic tools in healthcare.