EFFICIENT NETB3-POWERED DEEP LEARNING ARCHITECTURE FOR AUTOMATED SKIN DISEASE IDENTIFICATION IN CLINICAL APPLICATIONS

Abstract

Skin diseases represent a major global healthcare challenge, affecting millions of individuals and often requiring timely diagnosis to prevent severe complications. Traditional dermatological diagnosis relies heavily on expert clinical assessment, which can be time-consuming, subjective, and inaccessible in resource-constrained regions. Recent advances in artificial intelligence and deep learning have demonstrated significant potential for automated skin disease detection using dermoscopic and clinical images. This study proposes an Advanced EfficientNetB3-Based Deep Transfer Learning Architecture for accurate and robust skin disease classification. The proposed framework leverages the EfficientNetB3 convolutional neural network as a feature extraction backbone, benefiting from its optimized compound scaling strategy that balances network depth, width, and resolution while maintaining computational efficiency. Transfer learning is employed by initializing the model with pre-trained ImageNet weights, enabling effective knowledge transfer and reducing training time. To enhance classification performance, advanced data preprocessing, image augmentation, batch normalization, dropout regularization, and fine-tuning strategies are incorporated. The architecture is evaluated on a benchmark skin disease image dataset containing multiple dermatological categories, including melanoma, eczema, psoriasis, acne, and benign lesions. Experimental results demonstrate superior classification accuracy, precision, recall, F1-score, and area under the ROC curve compared with conventional convolutional neural networks and existing transfer learning models. The proposed model achieves improved generalization capability while reducing computational complexity, making it suitable for real-world clinical deployment and mobile healthcare applications. The findings indicate that EfficientNetB3-based deep transfer learning can serve as an effective decision-support tool for dermatologists, facilitating early detection, accurate diagnosis, and improved patient outcomes in skin disease management.