Abstract

Skin cancer, a widespread and potentially fatal malignancy, originates from the abnormal proliferation of skin cells, emphasizing the significance of early detection for effective treatment. Deep learning (DL) techniques, notably convolutional neural networks (CNNs), offer promise in the analysis of skin lesion images. This research study aims to compare various Convolutional Neural Networks (CNNs) classifier models using DL for skin cancer detection. The primary objective is to evaluate the effectiveness and performance of these CNN models in accurately classifying the type of skin cancer. To address the class imbalance, the data was first augmented, after which it was normalized to scale pixel values. Throughout the research, six CNN models were evaluated under different configuration settings, and the results were presented through tabulations, providing insights into their comparative performance. These outcomes contribute to the field of skin cancer detection by shedding light on the effectiveness of DL models. The analysis conducted in this study concludes that the highest results were recorded at an accuracy of 99.73% by ResNet50 using the Adam optimizer at a learning rate of 10^-3.

Authors

Sharmiladevi S; Shravan Venkatraman; Santhosh Malarvannan; Shriyans A; Swathisree S