<![CDATA[The utilization of Artificial Intelligence in the current era of technological development is increasingly popular, especially in the field of health. The increasing number of skin cancer cases globally is of particular concern today. Therefore, a classification model utilizing deep learning was developed to assist in the effective diagnosis process. However, data limitations and imbalances are often an issue in training skin cancer classification models. This research develops a skin cancer classification model using the Few-shot Learning approach with Prototypical Networks architecture and EfficientNet-B0 backbone. The research aims to develop an image-based skin cancer classification model and evaluate how effectively the model performs in classifying various types of skin lesions. Experimental results show that increasing k-shots has a positive impact on model accuracy. The best results were obtained in the 10-shot 15-query scheme with an accuracy value of 86.73% and supported by an ROC AUC value of 94%. This study proves that the few-shot learning approach with Prototypical Networks architecture and EfficientNet-B0 backbone is effective for skin cancer classification under limited dataset conditions. This model also has the potential to be an early diagnosis tool.]]>
Copyrights © 2025
