This paper presents a computational model for melanoma skin cancer identification using deep learning with transfer learning and explainable AI (XAI). The model utilizes the HAM10000 dataset, which contains 10,000 images of skin lesions, including melanoma and nevus. The study introduces a mask-guided technique that employs a U2-Net segmentation module to generate masks for image segmentation. Both convolutional neural networks (CNNs) and vision transformers (ViT) are used for classification, with the CNN-based approach achieving an accuracy of 98.37% using the Xception model, and the ViT-based approach achieving high accuracy, sensitivity, and specificity. The study also applies Grad-CAM and Grad-CAM++ to generate heatmaps for model explainability, which help to visualize the contribution of each input region to the classification outcome. The proposed model is developed as a web application for real-time use by medical practitioners. The system usability study score of 86.87% indicates the usefulness of the proposed solution. The main contributions of this work include applying data augmentation to address imbalanced datasets, training the U2-Net model to generate segmentation masks, conducting a comparative study of different CNNs and ViT-based models, identifying model performance using hyperparameter tuning, enhancing ViT performance for fine-grained visual categorization, providing integrability for fine-tuning on ViT-based backbones, improving system trustworthiness using explainability methods, and developing a web application for the proposed model. The study also evaluates the model using intersection over union (IOU) and other qualitative metrics. The results show that the proposed model achieves high accuracy and performance in melanoma classification, with the CNN-based model achieving the highest accuracy. The study also demonstrates the effectiveness of the mask-guided technique in improving model performance and explainability.This paper presents a computational model for melanoma skin cancer identification using deep learning with transfer learning and explainable AI (XAI). The model utilizes the HAM10000 dataset, which contains 10,000 images of skin lesions, including melanoma and nevus. The study introduces a mask-guided technique that employs a U2-Net segmentation module to generate masks for image segmentation. Both convolutional neural networks (CNNs) and vision transformers (ViT) are used for classification, with the CNN-based approach achieving an accuracy of 98.37% using the Xception model, and the ViT-based approach achieving high accuracy, sensitivity, and specificity. The study also applies Grad-CAM and Grad-CAM++ to generate heatmaps for model explainability, which help to visualize the contribution of each input region to the classification outcome. The proposed model is developed as a web application for real-time use by medical practitioners. The system usability study score of 86.87% indicates the usefulness of the proposed solution. The main contributions of this work include applying data augmentation to address imbalanced datasets, training the U2-Net model to generate segmentation masks, conducting a comparative study of different CNNs and ViT-based models, identifying model performance using hyperparameter tuning, enhancing ViT performance for fine-grained visual categorization, providing integrability for fine-tuning on ViT-based backbones, improving system trustworthiness using explainability methods, and developing a web application for the proposed model. The study also evaluates the model using intersection over union (IOU) and other qualitative metrics. The results show that the proposed model achieves high accuracy and performance in melanoma classification, with the CNN-based model achieving the highest accuracy. The study also demonstrates the effectiveness of the mask-guided technique in improving model performance and explainability.