Skin Disease Classification using Pretrained Transfer Learning Models

Cancer is a major global health concern, with skin cancer accounting for a sizable proportion of cases globally. Timely identification plays a crucial role in ensuring successful therapy; nevertheless, discerning between harmless and cancerous skin abnormalities poses a significant challenge owing to their comparable visual attributes. This frequently results in needless biopsies, emphasizing the need for better diagnostic tools. In response, here provided an improved image classification model intended to be used as a preliminary screening tool by dermatologists. The models are trained on the HAM10000 dataset, combines Transfer learning(TL) with ten pre-trained models and uses data augmentation approaches to improve performance. Here ResNet50, ResNet50V2 achieved 99.6%,99.7% training accuracy, and 98.5%,98.2% validation accuracy respectively. This technique can be helpful clinical decision support system that helps dermatologists with diagnostics and may reduce the need for needless biopsies.