Comparative Analysis of Machine-Learning Algorithms for Accurate Diagnosis of Lung Diseases Using Chest X-ray Images: A Study on Balanced and Unbalanced Data on Segmented and Unsegmented Images

Abstract

The study focused on the accurate diagnosis of lung diseases, considering the high number of lung diseaserelated deaths in the world. Chest x-ray images were used as they are a cost-effective and widely available diagnostic tool. Eight different machine learning algorithms were evaluated: Logistic Regression, Naive Bayes, k-Nearest Neighbors (kNN), Decision Tree, Random Forest, Support Vector Machine (SVM), Ridge, and Least Absolute Shrinkage and Selection Operator (LASSO). The study evaluated balanced and imbalanced datasets and looked at both segmented and unsegmented chest x-ray images. COVID-19, pneumonia, normal, and others were the four classes that were used in the investigation. Prior to attribute reduction, Decision Tree and Random Forest performed well on the balanced dataset, obtaining 74% test accuracy and 92% training accuracy. SVM functioned well as well, obtaining a 74% test accuracy. Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) are two attribute reduction approaches that were applied. Decision Trees and Random Forests were able to attain the maximum training accuracy of 92%, while SVM was able to retain a test accuracy of 74% after attribute reduction. The findings also imply that some algorithms' performance may be enhanced by attribute reduction methods like PCA and LDA. For imbalanced data, Random Forest and SVM perform the best in terms of balanced accuracy of 80%. However, further research and experimentation may be needed to optimize the models and explore other potential algorithms or techniques.