Classifying Multiple Skin Infections using Machine Learning

This paper presents a deep learning approach for the classification of multiple skin diseases, leveraging the prowess of Convolutional Neural Networks (CNNs) and a dataset rich in dermatological diversity. We utilize a model architecture comprising a pre-trained feature extraction layer coupled with a densely-connected softmax layer to classify images into one of nine distinct skin disease classes. The model was optimized using the Adam optimizer and trained across various demographics to ensure robust generalizability. During training over 10 epochs, the model demonstrated a remarkable improvement in performance, with accuracy soaring from 49.19% to 99.7% and loss decreasing from 1.4764 to 0.0621. A detailed classification report confirms the model’s efficacy, with precision, recall, and F1-score all above 0.91, and an overall accuracy reaching 97%. These results underscore the model’s potential in enhancing diagnostic accuracy for skin disease identification.