Research Article
Lightweight ShuffleNetV2-Based Deep Learning Framework for Accurate and Interpretable Skin Cancer Classification
@INPROCEEDINGS{10.4108/eai.19-12-2025.2366209, author={Krishna Kumar and Avantika Mishra and Anushka Singh and Ananya Srivastava}, title={Lightweight ShuffleNetV2-Based Deep Learning Framework for Accurate and Interpretable Skin Cancer Classification}, proceedings={Proceedings of the 1st International Conference on Computational Techniques in Data Science, IICTDS 2025, 19-20 December 2025, Chandigarh, India}, publisher={EAI}, proceedings_a={IICTDS}, year={2026}, month={11}, keywords={Skin cancer classification Deep learning ShuffleNetV2 Dermoscopic images Grad-CAM Binary classification Explainable AI}, doi={10.4108/eai.19-12-2025.2366209} }- Krishna Kumar
Avantika Mishra
Anushka Singh
Ananya Srivastava
Year: 2026
Lightweight ShuffleNetV2-Based Deep Learning Framework for Accurate and Interpretable Skin Cancer Classification
IICTDS
EAI
DOI: 10.4108/eai.19-12-2025.2366209
Abstract
Early detection of skin cancer is crucial, yet histopathological diagnosis remains chal lenging due to the visual similarity of lesions. This work proposes an interpretable, lightweight diagnostic model using ShuffleNetV2 as the backbone for efficient computation. Preprocessed dermoscopic images were used to train the model with regularization techniques such as la bel smoothing, Mixup, and calibration to improve generalization. Interpretability was ensured through Grad-CAM visualizations. On an independent test set of 200 images, the model achieved 91.5% accuracy, a macro F1-score of 91.5%, ROC-AUC of 0.9675, MCC of 0.83, and an ECE of 0.0918. Grad-CAM confirmed that clinically relevant lesion regions were consistently high lighted. The proposed method outperformed several classical and state-of-the-art CNN models, demonstrating its potential for real-time dermatology applications.