A Comparative Analysis of Skin Cancer Detection using Shallow and Deep Learning Features

Since skin cancer is the most common type of cancer, it has an annual impact on millions of lives. Every year, almost 3.5 million people receive a skin cancer diagnosis in the United States alone. The likelihood of survival decreases substantially with the advancement of skin cancer. Even so, early detection of this particular cancer type is a tricky and expensive process. Melanoma is a form of skin cancer that affects a broader spectrum of people than other cancer types. Prompt identification and treatment of this type of tumor may reduce the likelihood of its spread to other parts of the body, where successful treatment and cure are possible. The introduction of machine learning and deep learning techniques paved the way for the implementation of an effective automated diagnosis platform that could assist doctors in predicting diseases more rapidly and help patients detect them with skill. The models at present incorporate either deep learning-based techniques to extract features from entire images or machine learning models that are restricted to feature selection. For feature extraction and classification, the suggested hybrid pre-trained convolutional neural network and machine learning classifiers are employed. The accuracy of the model is enhanced by this kind of approach. Here, the hybrid ResNet101 V2 is employed for feature extraction together with greedy stepwise and multiobjective evolutionary search. This fusion achieves a maximum accuracy of 95.5%, which is significantly better than prior studies that are depicted in the literature study.