Melanoma is a very harmful and disastrous form of skin cancer, has prompted a surge of interest in utilizing deep learning methods, specifically Convolutional Neural Network (CNN), for its early detection through image analysis. However, this endeavor faces challenges, including a scarcity of training data, similarities between various skin lesion types, and variations within the same lesion class, compounded by the need to fine-tune numerous parameters in existing methods. To address these issues, this study introduces an automated framework that leverages pre-trained deep CNN models to extract visual features from skin lesion images, followed by the application of classifiers to identify melanoma. While earlier research has employed pre-trained CNN architectures to extract features, a comprehensive analysis of multiple CNN models for melanoma classification has been lacking. The research demonstrates that Alex Net, in combination with a multi-layer perceptron (MLP), achieves the highest accuracy at 88.5%, Outperforming other CNN models and current cutting-edge techniques in the crucial field of melanoma determination.