Abstract:
Skin diseases are the most commonly recognized health issues worldwide and are associated with
an extensive global burden; therefore, an easy and efficient diagnosis of skin diseases would be
crucial to ensuring the appropriate and effective treatment for the patient. Although there are many
computerized methods for diagnosing skin diseases, the convolutional neural network has proven
superior to the conventional methods.in recent research remarkable accuracy is achieved but
accuracy should be improved. This paper classifies ten types of skin diseases: atopic Dermatitis,
acne vulgaris, basal cell carcinoma, healthy lichen planus, melanocytic nevi, melanoma,
onychomycosis, tinea capitis, and vascular tumor. The primary objective of this paper is to develop
a skin disease diagnosis model by comparing different CNN architectures and evaluating the
performance of these deep learning networks on skin lesion images. The dataset was collected
from the Goba referral hospital and labeled by professionals. the lesion classification is
implemented through transfer learning on four CNN architectures: ResNet50, DenseNet201,
MobileNetV2, and MobileNetV3. The dataset used for these experiments is a custom dataset of
about 21,000 images. The results show that DenseNet 201 and Mobile Net v3 perform best with
89.2% and 89.0% accuracy, respectively. The proposed work shows the various parameters,
including the accuracy of all four deep learning networks, which helped build an efficient
automated classification model for multiple skin Diseases
