EF-SwinNet: A Hybrid EfficientNet-Swin Transformer Model for Skin Cancer Classification
Skin cancer ranks as the most common type of cancer worldwide. This work presents EF-SwinNet, a hybrid architecture combining EfficientNet and Swin Transformer models, designed to classify skin lesions with high accuracy. The dataset used for training and testing is HAM10000, known for its class imbalance; several data augmentation methods were applied to alleviate this. The proposed hybrid model achieved an average accuracy of 98% and an F1 score of 96%. Grad-CAM was used to visualize which features the model learned to attend to. Performance was benchmarked against current state-of-the-art methods, highlighting improvements in skin cancer classification.
Recommended citation: L. Sarker, et al. (2024). “EF-SwinNet: A Hybrid EfficientNet-Swin Transformer Model for Skin Cancer Classification.” IEEE ICRPSET 2024. https://doi.org/10.1109/ICRPSET64863.2024.10955919