A DEEP LEARNING APPROACH TO SKIN CANCER CLASSIFICATION

Abstract

This paper talks about the worldwide pandemic of skin cancer and underlines the need of exact finding for anticipation. Due of early identification issues, dermatologists utilize deep learning, particularly CNNs. The review utilizes examining, dull razor, and autoencoder-put together division with respect to the MNIST: HAM10000 dataset of 10,015 examples and seven skin injury classes. Transfer learning utilizing DenseNet169 and ResNet50 models shows that DenseNet169's undersampling brings about amazing accuracy and F1measure, though ResNet50's oversampling succeeds in both. In view of the first paper's ResNet50, DenseNet161, and VGG16 (91% accuracy), this extension analyzes Xception, DenseNet201, and InceptionV3. The review proposes that various models and boundary change could upgrade skin cancer classification by 95%, working on symptomatic precision and protection endeavors.