The aim of this study is to improve the accuracy of skin cancer classification using Elastic GAN method based on HAM10000 dataset. This dataset consists of 10015 dermoscopic images of pigmented skin lesions in seven diagnostic categories. Due to the class imbalance in this dataset, conventional data augmentation methods cannot provide sufficient diversity. In this study, Elastic GAN is used to generate new and diverse samples by applying elastic changes that simulate tissue changes. This method is based on competitive generative networks that can produce realistic images. First, the images of the dataset are preprocessed and the class imbalance is reduced by incremental sampling and traditional methods. Then, Elastic GAN is trained and new samples are generated for minority classes. Finally, deep learning models (including modified DenseNet201 and MobileNet as feature extractors along with support vector machines) are trained on the augmented data. The results show that data augmentation with Elastic GAN significantly improves the classification accuracy compared to traditional data augmentation methods. Specifically, the proposed model achieved an accuracy of 95.5%, which is a 2% improvement over the case without data augmentation. Also, the sensitivity indices of 93.96% and specificity of 97.03% indicate the reliability of the model in diagnosing malignant and benign cancers. In conclusion, Elastic GAN, as a novel data augmentation method, can improve the performance of skin cancer classification by generating diverse and realistic artificial samples and provide an effective solution to face the challenge of unbalanced data in the medical field.