Comparative and Ensemble Approach for Lung Cancer Detection Using Deep Learning Models

This paper presents a comparative study and ensemble approach for detecting lung cancer using deep learning models. We used three different model which are ResNet50, DenseNet121, and custom CNN. It is used to classify CT scans image into benign, malignant, and normal classes. To improve the learning process, we have applied data augmentation and SMOTE for balancing the different classes. Each model was fine-tuned using transfer learning techniques and then combined with an ensemble method which averages their predictions through logistic regression. Our results show that DenseNet121 performed best according to the accuracy. While the ensemble strategy for further improved robustness and overall prediction accuracy. These findings suggest that combining different deep learning models can enhance the reliability of lung cancer detection systems and offering promise for earlier diagnosis and better patient outcomes.