Skin cancer is currently one of the most common cancers worldwide; its early-stage detection is crucial for effective treatment. This work, therefore, proposes an automated deep learning-based framework to support skin cancer diagnosis. The dataset undergoes pre-processing, a prerequisite to enhance data quality by minimizing noise, increasing contrast, and reducing skin lesion variations. In the proposed work, the EfficientNet architecture offers a fine balance between high accuracy and computational efficiency. Further performance gain emerges from an attention mechanism that enables the model to focus on important lesion features, thereby increasing its ability to distinguish between benign and malignant cases. The model trains and validates on the ISIC dataset, a widely accepted benchmark for skin cancer diagnosis. With an accuracy of 94.7%, the proposed framework demonstrates proficiency in handling various skin lesion types. Additional performance metrics, including precision, recall, F1-score, and AUC-ROC, comprehensively evaluate the system’s performance. The results show that the EfficientNet model with an attention mechanism significantly outperforms traditional methods in terms of diagnostic accuracy and reliability. This framework represents an essential tool in dermatological practice, as it enables earlier and more accurate diagnosis of cutaneous malignancies during skin work-ups, leading to improved clinical outcomes through timely treatment.