Explainable AI for Skin Cancer Classification Unlocking Insights with Grad-Cam and Grad-Cam++

Abstract

Skin cancer is a long-term global public health problem with increasing incidence and high mortality, especially for malignant forms like melanoma. Although non-invasive diagnostic methods exist, early detection still relies to a large extent on clinical experience, so researchers have been investigating the use of artificial intelligence (AI) to improve diagnostic precision and efficiency. In this study, we integrate MobileNetV2, a lightweight yet powerful convolutional neural network, into a skin cancer diagnostic pipeline using the ISIC 2024 dataset. The deep learning model achieves a very good degree of diagnostic accuracy with 97.7 % accuracy and 99.2 % recall in differentiating between benign and malignant lesions, hence reducing the risk of false negatives. A novel contribution is the integration of Gradient-weighted Class Activation Mapping (Grad-CAM) and its improved variant, Grad-CAM++, to alleviate the “black-box” aspect of AI models. Both techniques produce visual heatmaps that highlight the most relevant areas in an image, enabling clinicians to have an interpretable rationale for the model's predictions. Comparative studies show that GradCAM++ yields more accurate and clinically meaningful heatmaps, thereby enhancing healthcare professionals' trust in automated diagnosis. This combination of high diagnostic capability, computational efficiency, and improved interpretability foretells widespread clinical adoption and emphasizes the significant role that explainable AI can play in advancing the future of dermatologic medicine.