Lung Disease Detection Using Hybrid CNN‐XGBoost Classifier With New Loss Function

Lung disease is one of the fatal and common illnesses, affecting many people each year through various forms such as asthma, pneumonia, and lung obstruction. Numerous diagnostic methods have been proposed, most of which rely on machine learning algorithms or simple convolutional neural networks (CNNs) that often lack sufficient accuracy. Given the diversity of lung diseases and X‐ray images, accurate diagnosis and classification remain complex and challenging tasks. To address this, a hybrid algorithm combining CNN and Extreme Gradient Boosting (XGBoost) is introduced, where the CNN first processes the training data, and its output is passed to the XGBoost classifier to produce the final prediction. Additionally, a novel loss function has been designed to overcome the limitations of the conventional softmax loss—such as sensitivity, class imbalance constraints, and mismatches between intraclass and interclass distances. The CNN‐XGBoost hybrid is employed to mitigate the weaknesses each model exhibits when used independently. This proposed method was evaluated on the NIH CXR dataset, which includes 15 classes (14 disease classes and one healthy class), achieving a test accuracy of 97.92%, significantly outperforming previous approaches.