High-Accuracy Pneumonia Classification via Ensemble Learning on Chest X-ray Imagery

Abstract

Pneumonia continues to pose a substantial global health threat, necessitating rapid and precise diagnostic tools. The conventional manual assessment of Chest X-ray (CXR) images is time-intensive and susceptible to human error. This study introduces an automated machine learning approach that employs an ensemble learning strategy to achieve highly accurate pneumonia classification from CXR images. The comprehensive system operates through three primary phases: initial image pre-processing (involving grayscale conversion, resizing, and filtering for enhanced quality), robust feature extraction (utilizing the fusion of Histogram of Oriented Gradients (HOG) and Local Binary Pattern (LBP) descriptors), and sophisticated model training and classification. An ensemble model is trained, integrating the predictive power of Random Forest, Logistic Regression, and Extreme Gradient Boosting classifiers. Experimental validations, performed on a dedicated dataset comprising pneumonia and normal CXR images, unequivocally demonstrate that the proposed strategy achieves an impressive 97.50% overall classification accuracy, strongly supported by precision, recall, and F1-scores all at 97.50%. This superior performance, notably surpassing individual machine learning algorithms, underscores the profound efficacy of ensemble learning in delivering reliable and precise predictions for pneumonia diagnosis. Consequently, this automated methodology presents a valuable asset for medical professionals, aiding in the swift and accurate identification of pneumonia.