Advancements of Skin Cancer Classification using Transfer Learning Segmentation

Skin cancer poses a significant health risk, with genetic alterations, tobacco use, sunlight exposure, and eating disorders identified as primary causative factors. Early diagnosis is crucial for effective treatment and improved prognosis. This study employs a Convolutional Neural Network (CNN) framework utilizing ResNet101 and Transfer Learning to classify two tumor types and a non-cancer category using the HAM10000 dataset. Coupled with class-weighted loss and data augmentation techniques, this approach yields dependable, rapid, and cost-effective results. Dermatologists can benefit from early disease diagnosis. Experimental results demonstrate that the modified ResNet101 model achieves a 92% classification performance and average weighted accuracy in distinguishing afflicted skin images among seven categories.