Skin cancer ranks among the leading causes of deaths globally and the effective treatment of its patients is largely dependent on early detection and accurate diagnosis. In the recent years, deep learning techniques have the top-notch results for a lot of computer vision tasks, such as image and content classification. Within this research, we present the way for skin cancer detection through the use of vision transformer, a combination of a promising deep learning architecture and algorithms for image analysis. In this study, a publicly available dataset called the HAM10000 is employed; HAM10000 contains 10,015 images with skin lesions that were divided into two groups while rectifying the images’ area which was benign (6705 images) or eliminating these area which was malignant (3310 images). The high quality dermatoscopic images in this data set are expert annotated and have a high grade of detail. The model completeness is reached through the use of both normalization and augmentation which are techniques for enhancing the model generalization creating robustness. The model was for skin cancer. Thanks to the self-attention component, the model is able to identify complex spatial and long range interactions of the images thereby equipping the model to perform the task of feature mapping for the purpose of providing better classification. The SAM is used for treating the image parts affected by cancer.