EfficientNetB7-Based Deep Learning for Skin Lesion Classification

Skin cancer ranks among the most prevalent types of cancer worldwide. For the purpose of giving effective treatment to the patients, it has to be treated in its initial stages. This study introduces Skin Lesion Classification using deep learning techniques with EfficientNetB7. It is trained on the ISIC 2020 Challenge dataset. Data preprocessing included organizing image file paths and labels, followed by data augmentation with Image Data Generator. The EfficientNetB7 model has been pre-trained on ImageNet. It was fine-tuned with additional BatchNormalization, Dense, and Dropout layers. After five epochs of training, the model demonstrated high accuracy in classifying various skin lesions, as evidenced by the validation and test results. Our project's findings emphasize the model's potential as a dependable automated system for detecting skin cancer, assisting dermatologists in early diagnosis and treatment. To enhance our work, we can consider hyperparameter tuning, advanced data augmentation, and enlarging the dataset for enhanced model generalizability.