Skin disease classification using Deep learning and Ensemble methods

Skin disease detection and categorization is a critical area of medical diagnostic research, aiming to enhance the early and accurate diagnosis of skin disorders. The goal of this research is to utilize advanced Convolutional Neural Network (CNN) architectures to achieve high classification performance efficiently. A novel 10-class skin disease dataset was created by collecting data from Kaggle, online sources, and various DermNet resources. To ensure data consistency and robustness, preprocessing techniques such as segmentation, normalization, augmentation, resizing, and class balancing were applied. This study employs three advanced CNN models: ResNet50, DenseNet121, and EfficientNetB2, alongside ensemble techniques such as stacking and soft voting. The proposed models demonstrated exceptional accuracy, achieving 98% with the HAM10000 dataset and 97.2% with the final ensemble stacking model. These results underscore the effectiveness of incorporating diverse datasets and ensemble methods to improve classification performance. The rapid, precise, and efficient methodology developed in this research holds significant promise for real-world diagnostic applications, promoting early detection and improved patient outcomes while reducing dependency on manual diagnosis.