Skin Lesion Classification using Transfer Learning

Skin Lesion also termed Skin Cancer has continuously recorded a high rate of mortality due to the ever-growing population, global warming, and various gases or pollution present in the atmosphere. Skin Lesions or Skin Cancer can be a horrifying way to die if not diagnosed early. Mainly Skin Lesion like Melanoma has been proven to be lethal. The mortality rate can be reduced if the skin disease is diagnosed at an early stage. The advancements in the Deep Learning community have been able to provide a way to diagnose skin diseases early. In this paper, the usage of pre-trained image classification model EfficientNetB0 is the proposed model which is used to classify 7 types of skin disease derived from the HAM10000 skin lesion dataset with Data Augmentation to increase the accuracy and help Dermatologists to classify and diagnose Skin Cancer early so it can be treated and can also be a way to cut down the cost of diagnosis. This project’s training accuracy and validation accuracy came out to be 97.61% and 93.50%. The weighted average and macro average precision, recall, and f1-score were 95%, 94%, and 95%. This paper proposes 90.5% accuracy to detect the most invasive skin cancer which is Melanoma and can help Dermatologists as a Decision Support System in the diagnosis process and create an application for ease of use.