Lung diseases produce a high morbidity and mortality rate among millions of people-spectacular effects around the world. An instant and accurate diagnosis of such chronic illness is required to be able to treat the patient accordingly. For example, chronic lung diseases are described as long-term respiratory or thoracic diseases with significant actuality on overall lung function and health of the individual. These require prolonged medical therapy and can lead to more complex forms of management if not treated properly. Chest X-rays are useful non-invasive methods of detecting and classifying various lung pathology features; faster treatment could also improve patient outcomes. The publication outlines the full review on the generation and evaluation of deep learning models with the NIH Chest X-ray Dataset for the detection of a wide array of thoracic abnormalities. The NIH dataset pools 5,606 images annotated using methods derived from radiological reports to further develop strong models. Pre-trained models like VGG19, DenseNet121, and EfficientNet were developed along with traditional machine learning classifiers—XGBoost, KNN, SVM, and Random Forest, using weighted loss to tackle data imbalanced issues. The study offers insights into each illness class’s classification performance through a complete exploratory data analysis, preprocessing, model training, and evaluation. Regarding the VGG19 model, the highest accuracy was achieved by Random Forest at 95.21%, followed closely by XGBoost at 95.10%, KNN at 94.83%, and SVM at 93.29%. For the DenseNet121 model, Random Forest again demonstrated the highest accuracy at 94.65%, followed by XGBoost at 94.56%, KNN at 94.38%, and SVM at 92.53%. For the EfficientNetB0 model, Random Forest achieved an accuracy of 95.16%, followed by XGBoost at 94.78%, KNN at 94.57%, and SVM at 91.0%. Among these models, Random Forest showed the best performance for VGG19, DenseNet121, and EfficientNetB0.