Skin melanoma is one of the most death-defying form of melanoma, stemming from DNA damage. Rapid identification with accuracy in its earlier stages is a paramount requirement for diagnosis and survival. Techniques like ‘Dermatoscopes’ which analyze images are often considered as the most effective method for detecting skin cancer. However, image analysis can be very challenging due to several troublesome factors. In this context, a novel approach centered on dense and deep learning has been introduced to enable the identification of subtle distinctions between skin lesions at the pixel level. Herein, an advance method for skin lesion classification is introduced, employing Convolutional Neural Networks (CNNs) with two cutting-edge architectures, ResNet-50 and VGG-19. The method put forward includes fine-tuning both ResNet-50 and VGG-19 on an extensive dermatology dataset, followed by an ensemble learning approach to synergize their capabilities. The proposed method was assessed using 10000 images retrieved from the Melanoma Skin Cancer Dataset and demonstrate notable enhancements in accuracy, sensitivity, and specificity when contrasted with current advanced techniques. The findings illustrate the capability of the contemplated CNN-powered approach, utilizing ResNet-50 and VGG-19, to function as a dependable tool for automating the diagnosis of skin cancer, enabling earlier detection and improved disease management.