Skin Cancer Classification Using Transfer Learning With MobileOne: A Deep Learning Approach

Abstract

Skin cancer is common and a rising cause of death globally. Early detection using deep learning helps save lives by enabling early and effective treatment. In this study, we utilized a pre-trained lightweight model, MobileOne, to classify malignant or benign skin cancer images. We combined two publicly available Kaggle datasets (ISIC 2019–2020 Malignant or Benign and Skin Cancer: Malignant vs. Benign). The proposed pre-trained model MobileOne outperformed existing models like MobileNetV3, Xception Net, Inception V3, VGG16, Densenet-121, ResNet50, VGG19, ViT b16, and ViT b32, achieving accuracy of 92.61%, precision of 0.9276, recall of 0.9261, F1-score of 0.9261, and ROC of 0.98. The results suggest that combined datasets with the lightweight MobileOne model improve accuracy and offer potential for fast, low-latency mobile skin cancer diagnosis, especially in resource-limited and remote settings.