Skin Diseases Classification Using Transfer Learning Model: Case Study of Tropical Regions

Skin diseases are a prevalent concern in tropical regions, particularly in African countries, due to the exacerbation of conditions caused by hot climates and poor living conditions. Diagnosis typically relies on dermatologists' manual inspections, which face challenges due to a shortage of specialists and inadequate healthcare infrastructure. These limitations often lead to diagnostic delays, treatment errors, and severe health outcomes. To address these issues, this paper proposes an AI‐assisted preliminary screening tool designed to support clinical decision‐making, particularly in settings with limited dermatologist access. The system provides probabilistic classifications requiring professional clinical validation. The tool leverages advancements in artificial intelligence and deep learning. A convolutional neural network (CNN) based on the MobileNetV2 architecture is exploited for classifying skin diseases. The methodology involves data collection, image preprocessing, data augmentation, and transfer learning to optimize model performances. The model is trained on a custom dataset comprising 1560 images of seven prevalent skin diseases in tropical regions. It achieves notable performance metrics in development dataset evaluation (95% accuracy), with preliminary real‐world feasibility validation on 40 clinical images demonstrating 90% accuracy. Additionally, a mobile application has been developed to deploy the trained model, offering an offline and real time mean to identify potential skin conditions for professional evaluation on smartphones. This innovation provides an accessible healthcare solution for remote and underserved regions, bridging gaps in medical infrastructure and specialist availability.