Hybrid CNN Feature Fusion Model with Class Weighted Balancing and Explainable AI for Thoracic Diseases Classification

Pneumonia and COVID-19 are both thoracic diseases that mainly affect the lungs and make breathing difficult. Radiological imaging techniques, particularly chest X-rays (CXR), play a crucial role in identifying these conditions. Although some signs on the X-ray may look similar, pneumonia usually appears as a solid white area in one part of the lung, while COVID-19 often shows up as light, cloudy areas spread across both lungs. To detect and classify these respiratory diseases the proposed method uses a hybrid CNN that combines features from pre-trained VGG16 and MobileNet networks with global average pooling to capture multi-scale representations from chest X-rays. The model is trained using 4-fold cross-validation with class-weighted loss and data augmentation. As fusion combines the captured features from both networks, representing multi-scale, multi-level feature representations that improve performance, the model leverages K-Fold cross-validation, class-weighted training, and data augmentation to demonstrate generalization and robustness, while using pre-trained networks with global average pooling and visualizing intermediate features to provide both efficient learning and preliminary explainability. We evaluated the validity of the proposed approach on two datasets with varying class distributions, achieving 97% accuracy for binary classification and 98.25% for multi-class classification.