Publication

Abstract : Recognizing the signs and symptoms of various skin disorders helps make a clinical diagnosis and administering appropriate therapy much easier. Based on a multi-stage feature-processing technique, this study provides a multi-stage feature-processing, and machine learning classifier ensembles approach to skin disease identification. The classifier includes both the Voting Classifier and the Transfer Learning application. The architecture proposed is based on the concept of transfer learning and includes a large number of previously trained deep convolutional neural networks. It is also placed between the post-transfer learning process and the feature selection strategy, assisting us in automatically determining which features are most relevant for prediction and hence which features are most significant for prediction. Following the extraction of the deep features, the classifiers do data analysis. After a large number of classifiers have been deployed, the voting classifier makes predictions about the image`s content. This set of deep features is fed into a huge number of machine learning classifiers, which are then used to forecast the final result. The research uses ISIC`s publicly available skin image datasets to better understand how different types of pre-trained models, machine learning classifiers, and deep feature extractors differ. A study of big skin datasets shows that combining deep features with a voting classifier improves overall performance in the skin classification job.