Identification of Skin Cancer Using Deep Learning

Early detection of skin cancer is essential for improving patient survival rates and treatment outcomes. This study investigates the performance of Convolutional Neural Networks (CNNs) architectures s in skin cancer classification and addresses the challenges posed by class imbalance through a hybrid sampling strategy. The proposed balancing techniques significantly improved model sensitivity to underrepresented malignant lesions. Experimental results show that ConvNeXtLarge achieved the highest performance, with a recall of 82.76%. With a resampling method applied, the recall of cancerous classes, particularly melanoma, increased across all models, demonstrating notable improvements over the imbalanced dataset.