Transfer Learning-Based Skin Cancer Classification Using MobileNetV2, ResNetV2, and EfficientNetV2

Skin cancer, a menacing ailment worldwide, poses grave threats due to its often incurable nature. Employing deep learning techniques can significantly enhance skin cancer classification accuracy, thereby aiding early detection and effective treatment. This study conducts a comparative analysis of three transfer learning neural networks—EfficientNet v2, MobileNet v2, and ResNet 50 v2—using the HAM10000 dataset with nearly 10,000 images of six distinct skin lesion types. MobileNet V2 and EfficientNet V2 outshine ResNet50 V2 in terms of average precision, recall, and f1-score, demonstrating their efficacy in skin cancer classification. Notably, MobileNet V2 excels with a 78% average precision, recall, and f1-score, outperforming ResNet's 68%. This research establishes MobileNet V2 as the optimal model for skin cancer classification among the three considered.