Enhanced Skin Cancer Classification using EfficientNetV2 with CBAM and Test-Time

Skin cancer is among the most common types of cancer worldwide, and early detection significantly increases treatment success. Deep learning approaches have shown great promise in automatic skin lesion classification. This paper presents an enhanced method for skin cancer classification using the EfficientNetV2 architecture augmented with Convolutional Block Attention Module (CBAM) and Test-Time Augmentation (TTA). EfficientNetV2 provides a scalable and optimized feature extraction backbone, while CBAM improves feature representation by focusing on salient spatial and channel-wise features. TTA boosts prediction robustness during inference. Experiments conducted on the ISIC 2018 dataset demonstrate that our approach achieves superior performance compared to standard baselines, with improvements in accuracy, F1-score, and AUC. These results suggest that our method can serve as a valuable aid in clinical decision-making.