Identify Skin Disorder Using Dermoscopic Analysis Through Federated Learning

Abstract

Skin disorders affect millions of individuals worldwide, necessitating accurate and early diagnosis for effective treatment. Traditional machine learning-based medical image classification often requires centralized data storage, raising concerns about data privacy and security. To address this issue, this project implements Federated Learning (FL) for skin disorder detection using dermoscopic images. The proposed system utilizes YOLOv8 for object detection and CNN for classification. The HAM10000 dataset is used for training, with data distributed across two federated clients. Each client trains locally, and the trained model weights are aggregated at a central federated server, ensuring data privacy while enhancing model generalization. The CNN model achieves a global accuracy of 95%, outperforming the VGG19 model (65%). A Flask-based web application is deployed to enable real-time skin disorder detection, offering a scalable and secure AI-driven diagnostic solution