A study and evaluation of transfer learning–based deep learning models for common skin disease detection

The application of artificial intelligence in medical image–based diagnosis has attracted increasing attention, particularly in dermatology, where accurate and timely analysis of skin lesions is essential. This paper investigates and evaluates the performance of several transfer learning–based deep learning models for the automatic detection of common skin diseases. Specifically, three representative convolutional neural network architectures—ResNet50, Inception, and VGG19—are comparatively analyzed. All models are trained and evaluated on the same clinical skin image dataset, employing a unified preprocessing pipeline and data augmentation techniques to enhance generalization performance. The models are assessed using Accuracy, Precision, Recall, and F1-score as evaluation metrics. Experimental results demonstrate that VGG19 outperforms the other architectures, achieving an accuracy of 94.42%, with Precision, Recall, and F1-score all exceeding 94%, indicating stable and well-balanced classification performance. In contrast, ResNet50 and Inception achieve accuracies of 58.37% and 89.7%, respectively, reflecting their comparatively limited effectiveness under the same experimental conditions. These findings highlight the suitability of VGG19 for skin disease detection tasks in scenarios with limited data and provide practical insights for selecting appropriate transfer learning models in real-world dermatological decision-support systems.