| Skin cancer is one of the most dangerous forms of cancer and can be lethal. In general an early diagnosis in the preliminary stages can significantly determine the probability of fully recovering. Nonetheless, early detection of skin cancer is an arduous and expensive process. Although this type of cancer is visible, this does not simplify the diagnosis, as cancerous tumours look extremely similarity to normal lesions. Examining all pigmented skin lesions via surgical treatments causes significant soreness and scarring. Consequently, there is a need for an automatic and painless skin cancer detection system with high accuracy. Recently, Machine Learning (ML) and Deep Learning (DL) have demonstrated promising results in prediction and classification, skin cancer detection has been performed exceptionally well by them. This paper compares the effectiveness of several DL models which tackle the problem of automatic skin cancer detection using pre-trained models of Convolutional Neural Networks (CNNs), namely ResNetv2, VGG16, EfficientNet-B5, and EfficientNet-B7. These are compared with a ML model, namely the Support Vector Machine (SVM), in order to determine whether or not the examined skin sample is cancerous. The results show that the four CNN models outperform the SVM in accuracy, precision, recall and F1-score, especially EfficientNet-B7 provides the highest F1-score to reach 84.22%. |
*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.