Skin cancer is a widespread and potentially fatal condition that necessitates precise and prompt diagnosis for effective treatment. Recent advancements in deep learning have yielded promising outcomes in automating the classification and localization of skin cancers. This study offers an in-depth analysis of skin cancer classification and localization utilizing the MobileNet deep learning model on a substantial dataset encompassing 81 categories, each containing 400 images sized at 64 × 64 pixels. The MobileNet model’s performance was assessed through various metrics, including precision, recall, F1-score, and time efficiency. The model demonstrated exceptional results, achieving a training accuracy of 99.94%, a validation accuracy of 99.12%, and a testing accuracy of 99.19%. The F1-score, recall, and precision were recorded at 99.18%, 99.19%, and 99.20%, respectively. Additionally, the model exhibited efficient time performance, with an inference time of 1.6 s. The Receiver Operating Characteristic (ROC) curve analysis further confirmed the model's robustness, presenting an area under the curve (AUC) of 100%. These results suggest that the proposed MobileNet-based approach effectively classifies and localizes skin cancers, delivering accurate and efficient outcomes on a diverse and extensive dataset. The findings of this research contribute to the enhancement of skin cancer diagnosis and treatment, providing healthcare professionals with a dependable and automated tool. The study underscores the potential of deep learning models in assisting dermatologists and clinicians in making precise and timely decisions, thereby improving patient outcomes and potentially saving lives.