Identify Skin Disorder Using Dermoscopic Analysis Through Federated Learning

Skin problems are common health issues around the world, and effective treatment frequently requires an accurate and prompt diagnosis. The objective of this work is to use federated learning to enable dermoscopic analysis for the identification of skin problems. Federated learning reduces the data sharing and privacy violations by decentralizing model training across several collaborating institutions, hence ensuring data security and privacy. To improve data privacy, the weights of the skin condition are kept in both the local client and the global client (federated server). Consequently, the Federated server balances the server for both Local clients. The YOLO8 approach is used to extract the features of the skin condition in order to classify it. The HAM10000 dataset is used to train the VGG19 and Convolutional Neural Networks (CNN) models. The accuracy rates of the local servers is 93% and the global server is 94%.