Automatic Diagnosis of Skin Diseases with Convolutional Neural Networks on Multi-Class Visual Data

Abstract

The automatic diagnosis of skin diseases is of great importance, especially in cases requiring early detection, as it accelerates clinical processes and reduces the margin of error. In this study, a classification model based on Convolutional Neural Network (CNN) architectures was developed on a multi-class visual dataset containing three different skin disease categories. To enhance the model’s performance, data augmentation techniques were applied, and the images were resized to 224×224 pixels. Using a transfer learning approach, the model was trained with preprocessing ResNet-18, AlexNet, and DenseNet-201 architectures. The hyperparameters used during the training process were carefully selected, and the model's training and validation accuracies were monitored. According to the results obtained, the ResNet-18 model demonstrated strong performance with an accuracy of 87.19% on the test set. These findings indicate that deep learning-based architectures can be effectively applied in the multi-class diagnosis of skin diseases.