<p>Lung cancer is one of the most prevalent malignancies and remains the leading cause of cancer-related deaths worldwide due to its late diagnosis and high mortality rate. This paper proposed deep learning techniques, especially transfer learning and hybrid methods, for automating the detection of lung cancer. At the initial stage, the dataset underwent preprocessing, including resizing and normalization through rescaling. Afterward, data augmentation techniques such as flipping, rotation, zooming, and contrast enhancement were employed to reduce overfitting and improve model robustness. Then, Error Level Analysis (ELA) was applied as an auxiliary preprocessing step to highlight localized intensity variations and compression artifacts that are often imperceptible in raw images. This study investigated deep learning techniques that can effectively identify different classes of lung cancer. First, we utilized transfer learning with pre-trained models, which are EfficientNetB0, EfficientNetV2B1, and ResNet50. Finally, we used hybrid models, which are CNN + EfficientNetB0 and CNN + EfficientNetV2B1. The models were evaluated under multiple classification settings to assess their performance and robustness. Results show that EfficientNetB0 achieved the highest accuracy of 93.28% under standard classification. Among hybrid models, CNN + EfficientNetV2B1 performed best with 96.70% accuracy. Additional experimental settings yielded higher performance scores (up to 99.97%); however, these gains primarily reflect optimization benefits rather than clinically meaningful discrimination between cancer subtypes. These findings indicate that the proposed models can accurately identify and classify lung cancer across different evaluation settings. The models we created can help pathologists, especially those with limited resources. Finally, these models can effectively identify and categorize lung cancer.</p>