<![CDATA[Skin diseases are a global health problem affecting more than 900 million people annually, with a prevalence of 15–25% in primary healthcare visits in Indonesia. Limited access to dermatologists and the concentration of 70% of specialists in urban areas often lead to delayed diagnoses. To address this issue, this study develops a skin disease detection system based on deep learning using the MobileNetV3 architecture, focusing on computational efficiency on mobile devices and improved accuracy through knowledge distillation techniques. The dataset consists of four categories of skin diseases collected independently, with the model trained using transfer learning and fine-tuning, and further optimized with knowledge distillation to enhance performance without increasing complexity. Evaluation results show excellent performance with an overall accuracy of 97%, surpassing the initial target of >85%. The average precision, recall, and f1-score reach 0.97, demonstrating consistent performance across all categories. In particular, the ringworm class achieved 100% recall, while other classes reached values above 93%. The research outputs include a well trained MobileNetV3 model for high accuracy skin disease classification and a scientific publication on model optimization. This system is expected to provide an affordable and accessible diagnostic support solution, particularly for healthcare workers and communities in underserved areas.]]>
Copyrights © 2025
