Enhancing Skin Disease Diagnosis: A Hybrid Approach Combining Vision Transformer and Feature Selection Techniques.

Skin disease diagnosis is essential for timely treatment and improved patient care, as skin conditions can indicate more serious health issues. Accurate diagnosis is crucial to prevent complications and enhance care quality. However, challenges like class imbalance and high-dimensional data often limit the performance of diagnostic models. Many existing methods struggle with accuracy and generalization across datasets. In this study, we propose an improved approach for skin disease diagnosis using the Vision Transformer (ViT) for feature extraction (FE) combined with feature selection (FS) methods such as Principal Component Analysis (PCA), Recursive Feature Elimination (RFE), Chi-square (χ2), and Mutual Information (MI). Machine learning classifiers, including Support Vector Machines (SVM), Extreme Gradient Boosting (XGBoost), and Random Forest (RF), were applied to enhance performance. The model was evaluated on the Skin Disease Classification Dataset (SDCD) and the HAM10000 dataset. Our results demonstrate significant improvements, with PCA and SVM achieving 99.89% accuracy on the SDCD dataset. Other FS methods, including Chi-square, MI, and RFE with XGBoost, also achieved accuracies above 99.85%. This study is the first to use the SDCD dataset for skin disease classification, establishing a strong baseline for future research. On the HAM10000 dataset, our approach achieved 99.01% accuracy using MI and RFE with XGBoost, the highest result among all tested FS methods and classifiers, outperforming previous work by 0.82%. These findings provide a robust and effective solution for automated skin disease classification, addressing gaps in prior research and offering improved accuracy and practicality for medical diagnostics.