Enhancing Lung Cancer Classification with Ensemble Deep Learning Models

This study creates ensemble models from several pretrained models to classify 3 classes of lung cancer, including benign, malignant, and normal. The pretrained models that we use are EfficientNetV2B0, ResNet50V2, and DenseNet121, which were trained separately and then integrated using the average ensemble technique to improve prediction accuracy. The model training process is carried out by utilizing transfer learning, where the basic layers of the previously trained model on the ImageNet dataset are frozen to maintain previously acquired knowledge. New fully connected layers are then added and trained to adapt the model to the specific classification task of lung cancer. In addition, the fine-tuning process is carried out by reopening some of the frozen layers to further refine the model’s performance on the lung cancer dataset that we use. The results of the study show that the developed ensemble models produces higher accuracy compared to individual models. With the ensemble models that have been built, it is hoped that it can be an effective tool in helping to classify lung cancer quickly, accurately, and can contribute to increasing the chances of patient recovery.