A Comparative Analysis of Various Transfer Learning Approaches Skin Cancer Detection
Linda K Ashim, Nimisha Suresh, C V Prasannakumar
Abstract
Skin cancer is an unusual tumor of skin cells. It usually develops in places exposed to direct sunlight, but it can also form in places normally not exposed to direct sunlight. The two main categories of skin cancer are defined by the cells involved. Simple resection, microscopic Mohs surgery, curettage and electrode therapy, and cryosurgery can be used to treat skin cancer. A cloud-based architecture using deep learning algorithms in key implementations is used to build models that can predict skin cancer more accurately. However, one of the main problems is the limited availability of microscopic images for training models. In order to overcome this difficulty, a transfer learning method is proposed. This article provides a comparative analysis of different transfer learning models such as Densenet, Xception, VGG16, EfficientNet, Resnet. All models are trained on Kaggle skin cancer malignant vs non benign dataset. Achieves 88.61%, 78.41%,81.94%, 78.44% and 72.37 % accuracy respectively.