<![CDATA[Skin disease monitoring among fishermen in remote coastal areas remains challenging due to limited access to medical facilities and delayed clinical assessment. This study proposes MobiDerm-Fisher, a transfer learning-based skin disease classification framework utilizing the MobileNetV3Large architecture combined with cost-sensitive learning to address class imbalance. The model was trained on 4,564 dermoscopic images from the DermNet dataset covering five classes: Psoriasis, Eczema, Atopic Dermatitis, Poison Ivy, and Urticaria, using a stratified 80:20 train–test split. Class weighting was integrated into the categorical cross-entropy loss function to ensure balanced learning across minority and majority classes. Performance evaluation was conducted using accuracy, precision, recall, F1-score, and confusion matrix analysis. The proposed model achieved a validation accuracy of 97.1% and a macro-average F1-score of 0.961, demonstrating consistent classification performance across all categories, including minority classes. The trained model was deployed into a Flask-based web application for real-time inference in low-resource environments. This study contributes to Informatics by presenting a lightweight and imbalance-aware deep learning framework optimized for mobile health deployment in resource-constrained maritime communities.]]>
Copyrights © 2026
