Abstract
The research is trying to move forward skin disease classification using a dataset consisting of 19,500 images of 23 distinct skin conditions retrieved from dermatology-pictures-skin-disease-pictures. The trial relies on ResNet50 structure, with a remarkable 97% clinical interpretation success. The images in JPEG format with different resolutions, at the beginning of processing, were divided into two groups and their size was changed in addition to standardization. Time was devoted to intensive training covering tuning of hyperparameters as well as growing data variety. This research work contributes to the current knowledge on skin lesion classification, evaluating ResNet50 features that identify different skin problems. Everything about the high accuracy indicates that deep learning has a huge potential in dermatology. The paper herein presents both the model's strengths and weaknesses to set the path for future developments. This study is the milestone toward the automated skin diseases diagnosis and also the development of image analysis technologies.