Explainable Multi-Model Lung Disease Classification from Chest X-Rays: A Full-Team Comparative Study

Chest X-ray imaging is one of the most accessible diagnostic tools for pulmonary assessment, but manual interpretation can be time-consuming and affected by visual similarity between disease patterns. This project presents a full-team experimental framework for nine-class lung disease classification using deep learning, transfer learning, hybrid machine learning, ensemble learning, and explainable artificial intelligence (XAI). Twelve individual model pipelines and several ensembles were evaluated. The team used Grad-CAM, LIME, SHAP, Integrated Gradients, Occlusion Sensitivity, central focus score, and heatmap entropy to examine model behavior. VGG16+SVM achieved the strongest reported individual result at 99.21% test accuracy. Omar Medhat’s weighted test-time augmentation ensemble achieved 97.63% accuracy and 97.86% macro F1-score. These results are experimental and require external validation before clinical use.