Deep Learning Model for Image-Based Skin Disease Classification

Abstract

Skin diseases are among the most prevalent health problems worldwide and require accurate diagnosis to ensure proper treatment. However, manual diagnosis is often time-consuming and highly dependent on the expertise of medical professionals, which may lead to misclassification and delayed treatment. Although many deep learning models have been proposed for skin disease classification, comprehensive comparisons between recent CNN architectures and the latest YOLO-based classification models remain limited, particularly on multi-class skin disease datasets. Therefore, this study aims to develop a computer-based system for the automatic classification of various skin diseases using deep learning approaches. The proposed method utilizes Convolutional Neural Network (CNN) and You Only Look Once (YOLO) architectures to perform feature extraction and classification on dermatological images. The dataset consists of 14,276 images across 18 skin disease classes, which are processed through image preprocessing, data augmentation, and model training stages to improve generalization and performance. Several pre-trained models, including MobileNetV4, ResNet50V2, EfficientNetV2-S, YOLOv11, and YOLOv26, were implemented and evaluated using accuracy, precision, recall, F1-score, and confusion matrix analysis. Experimental results show that the YOLOv26 model achieved the best overall performance, obtaining an F1-score, Precision, and Recall of 80%, outperforming the other evaluated architectures in both classification capability and training stability. These findings demonstrate that modern YOLO-based classification models provide a competitive solution for multi-class skin disease recognition and offer a comprehensive benchmark that may support AI-assisted dermatological diagnosis, particularly in healthcare environments with limited access to dermatology specialists.