A Melanoma Skin Cancer Detection and Classification System Using Deep Transfer Learning

Melanoma is one of the most lethal forms of skin cancer, and early detection is critical for effective treatment and improved survival rates. Traditional diagnosis relies on manual visual inspection of suspicious skin lesions, a process that is time-consuming and subject to human error. This study presents an automated Melanoma Detection System that leverages optimized deep transfer learning to reduce diagnosis time while maintaining high reliability. The proposed method employs a fine-tuned VGG16 model with data augmentation, dropout regularization, and adaptive learning rate optimization, trained on a combined dataset consisting of the Melanoma Skin Cancer Dataset (10,000 images) and the SIIM-ISIC Melanoma Classification dataset. The proposed technique achieve 89% accuracy, 89% precision, 92% recall, and 90.4% F1-score, outperforming baseline CNN and pre-trained architectures. The findings demonstrate that the integration of transfer learning with targeted optimization strategies can significantly improve early detection of melanoma, providing a robust and generalized solution for clinical support.