Pulmonary health challenges like pneumonia, COVID-19 as well as tuberculosis pose significant challenges to global health, each requiring targeted approaches for management and treatment. Pneumonia, often caused by bacterial, viral, or fungal infections, leads to lung inflammation and impaired breathing. COVID-19, driven by the SARS-CoV-2 virus, is marked by its rapid transmission and severe impacts on vulnerable populations. Tuberculosis, attributed to Mycobacterium tuberculosis, is still a significant global health challenge, particularly in low-resource settings. Improved treatment outcomes depend on the precise and prompt diagnosis of these conditions. A novel framework employing transfer learning and attention mechanism for predicting pneumonia, COVID-19, and tuberculosis is examined in this research. The proposed system integrates a robust preprocessing pipeline, incorporating Principal Component Analysis (PCA) to optimize feature selection and reduce computational complexity. PCA efficiently maps high-dimensional data to a compact form, ensuring key information is preserved while enhancing the Knowledge acquisition process’s speed and accuracy. The model leverages the EfficientNetB0 architecture, a state-of-the-art convolutional neural network, and attention mechanism, achieving an impressive classification accuracy of 99%. These results underscore the framework’s potential to deliver reliable diagnostic support. By combining PCA’s dimensionality reduction capabilities with the powerful feature extraction of EfficientNetB0 and attention mechanism, this study introduces a scalable and effective solution for the early detection of pulmonary diseases. This approach aims to support clinicians in making accurate diagnoses and improving healthcare outcomes.