Multi-Class Skin Disease Detection Using Deep Learning Hybrid Method

This paper aims to exploit deep learning techniques to make skin diagnostic processes more efficient, delivering higher accuracy and consistent predictions to assist dermatologists in making informed decisions and mitigating diagnostic errors. In this paper, a hybrid technique is exploited where two pre-trained models namely MobileNet and EfficientNetB0 have been integrated and Transfer/Ensemble Learning have been exploited. High quality images collected in two public datasets were considered and they were divided into three classes: eczema, scabies, and healthy skin. Multiple preprocessing steps have been applied, e.g., data augmentation, to make the model robust and more generalized. The developed model achieved acceptable results in terms of accuracy, precision, and recall associated to training and validation phases, showing artificial intelligence's ability to take dermatological diagnostics to the next level and helpful to initial assessment of skin particularly in low-resource communities.