Vol 9 No 1 (March 2027): OPEN
Title: Hierarchical Ensemble Learning for Automatic Skin Disease Classification Using Dermoscopy Images
Author(s): T. Sai Harshitha, M. Suneetha
Abstract: Skin illnesses, especially skin cancer, are growing increasingly widespread and require early diagnosis, making them a global health issue. Traditional diagnostic methods take time and require clinical expertise. This work proposes a hierarchical DL technique to automatically classify skin diseases using dermoscopic images to address these issues.The framework advises two-tiered classification. The first model uses binary classification to distinguish malignant from non-malignant skin lesions. A multi-class classification model classifies non-cancerous lesions into disease classes in the second stage. The hierarchical method aids diagnosis and classification.Ensemble learning with ResNet and ConvNeXt enhances prediction accuracy and robustness. Ensemble models represent features better and forecast better than individual models. Experimental results show strong binary and multi-classification precision and classification effect.Confidence-based decision techniques improve system reliability by recognizing incorrect predictions. Grad-CAM visualizes model decisions for medical staff.The algorithm is trained and evaluated on HAM10000 skin lesions. Experimental results showed that the suggested ensemble model had an AUC score of 0.9399 for cancer detection, 84% accuracy in Stage-1 binary classification, and 94% accuracy in Stage-2 multi-class classification. Ensemble forecasts were accurate and intelligible, surpassing individual models in precision, recall, and F1-score. These findings suggest that the framework can be used as a computer-aided diagnostic tool for teledermatology, clinical decision support, and skin disease diagnosis.
DOI: 10.36079/lamintang.ijortas-0901.1105















