In this work, an automated model was developed for the identification of skin cancer using deep learning and ensemble learning techniques. A dataset of 14,034 dermatoscopic and clinical images was used, which were preprocessed to enhance their quality. The model combines pre-trained architectures such as EfficientNetV2B0, DenseNet121, ResNet50, InceptionV3, Xception, and MobileNet through the stacking technique, complemented with Grad-CAM for the interpretation of diagnostic patterns. The results show a diagnostic accuracy of 90.7%, surpassing conventional methods and individual models. It is concluded that this system, integrated into desktop and web graphical interfaces, has the potential to significantly improve early skin cancer detection, facilitating its implementation in clinical settings.