Performance Comparison of Skin Cancer Detection and Classification using CNN

Skin cancer is a potentially lethal disorder that arises from the uncontrolled proliferation of aberrant cells seen in the skin’s outermost layer. Six out of every seven skin cancer-related deaths are caused by the deadly skin disease melanoma. This kind of cancer has already been identified by the application of machine learning to examine different imaging modalities and protein sequences. Hand-engineered characteristics are typically necessary for machine learning, which can be tedious and time-consuming. The scientific community worldwide has undertaken extensive effort to develop extremely accurate automated systems for skin cancer screening to prevent instances of false negative diagnoses. Furthermore, there is a high likelihood of false negative diagnoses in hospitals where dermatologists must diagnose several cases of skin cancer. More significantly, researchers’ attention has been greatly drawn to the identification of lesions in early diagnosis. By examining deep-learning-based convolutional neural networks (CNNs), artificial intelligence (AI)-based methods have aided in the early detection of skin cancer. For this we analyze the three deep learning models, including VGG16, ResNet50, DenseNet121 to identify and categorize seven distinct forms of skin cancer.Out of three models DenseNet121 has highest testing accuracy of 85%,with training accuracy 99%,0.67 weighted average precision, 0.66 weighted average recall , 0.66 weighted average f1-score.