Early diagnosis of skin diseases, especially melanoma and other skin cancer types, plays an important role in timely treatment and increases survival. Common diagnostic approaches depend on the knowledge and experience of expert clinicians, which is often laborious and susceptible to human error. GANs act as the data augmenters, CNNs are used for feature extraction, while XGBoost handles the classification. We utilize the publicly available ISIC dataset, comprising over 25,000 labeled dermoscopic images, to train and evaluate the proposed model. The GAN is employed to generate synthetic images, addressing the challenge of class imbalance in the dataset. The CNN extracts deep features from these augmented images, which are then passed to the XGBoost classifier for accurate skin disease classification. A comparative analysis is performed against other popular models, including CNN-only, GAN + CNN, and CNN + XGBoost. Our results show that the GAN \(\rightarrow \) CNN \(\rightarrow \) XGBoost model achieves an accuracy of 96.3%, outperforming the other models and demonstrating its potential for efficient and reliable skin disease detection.