Comparison between traditional machine learning algorithms and deep learning algorithms in pneumonia image classification

Chest X-ray (CXR) imaging is widely used to diagnose pneumonia and related diseases, and is critical for effective treatment and prevention. This paper aims to explore the employment of Visual Transformer (ViT) in pneumonia recognition, while using Logistic Regression (LR), Random Forest (RF), K-Nearest Neighbors (KNN) as classifier, or pure ViT classification, and compare the performance. Ultimately, the pure ViT classification model has the highest accuracy (84.3%), highest precision (80.2%), highest recall (99.7%), highest F1 score (0.867) and largest AUC value (0.97) on the training set, demonstrating accuracy and stability.