Enhancing Skin Disease Classification and Privacy Preservation through Federated Learning-Based Deep Learning

Human skin is a very sensitive part and is infected with various types of issues. Skin disease is one of them and 70% to 80% of the population is infected with one of the skin diseases. Expensive clinical screening and dermatological analysis are the only solutions to detect skin disease. Legacy classification methods are not appropriate to tackle the issues related to the color, texture, and privacy of the skin disease dataset. Both deep learning(DL) techniques and federated learning(FL)can be better solutions for classification and privacy issues related to skin datasets. The collaboration of both in the medical field can be a significant contribution. In this paper, federated learning-based deep learning was applied to a publicly available dataset consisting of 10 different skin diseases. Various deep learning models were applied, and privacy concerns were preserved with a federated learning approach. Here, the dataset was enhanced with several image augmentation strategies. In this research, several models were used and after observation, it was found that the InceptionNet outperformed and produced a better accuracy rate of 98.89%. In this research, complete experiments were set with varying client values and communication rounds.