Skin cancer is a significant health concern worldwide, necessitating effective early detection methods. This project investigates the efficacy of five prominent convolutional neural network (CNN) architectures VGG16, ResNet50, InceptionResNetV2, EfficientNetB4, and DenseNet121for skin cancer detection, utilizing the HAM10000 dataset. Through rigorous experimentation, each model’s performance is evaluated based on classification accuracy. Results reveal InceptionResNetV2 achieving the highest accuracy of 89%, followed closely by ResNet50 at 87%. EfficientNetB4 and DenseNet121 also demonstrate competitive accuracies of 86% and 81.62%, respectively, while VGG16 yields a comparatively lower accuracy of 78%. Additionally, to address the challenge of limited training data, generative adversarial networks (GANs) are employed for data augmentation. The GAN model, specifically the A skin lesion style-based GANs, is utilized to produce synthetic skin lesion images, augmenting the training dataset. The integration of generated images from the GAN model enhances the training of a deep transfer neural network classifier, contributing to improved classification performance. These findings underscore the potential of deep learning approaches in skin cancer diagnosis, emphasizing the importance of selecting appropriate CNN architectures and leveraging advanced techniques like GANs. The research contributes insights for clinicians and researchers in developing automated systems for early diagnosis, potentially enhancing patient outcomes and reducing healthcare burdens.