<![CDATA[Skin cancer is one of the most dangerous diseases that can become fatal if not detected accurately at an early stage. Conventional visual diagnosis performed by dermatologists still faces limitations due to subjectivity and dependence on clinical experience. Therefore, an objective, fast, and efficient diagnostic support system is required. Recent advances in artificial intelligence, particularly deep learning, enable the utilization of dermatoscopic images to assist in automated skin cancer detection. This study aims to implement and evaluate the performance of the Convolutional Neural Network (CNN) MobileNetV2 architecture for classifying skin cancer images into two classes, namely benign and malignant. The research adopts an experimental deep learning approach using the HAM10000 dataset, which consists of 10,015 dermatoscopic images. The data undergo preprocessing stages including image resizing, normalization, and data augmentation, followed by stratified splitting into training, validation, and testing sets. The MobileNetV2 model is trained using a transfer learning strategy through two stages, namely head training and fine-tuning. Experimental results show that the proposed model achieves an accuracy of 82.42%, with a malignant class precision of 64.04% and a recall of 19.93%. Although the model demonstrates strong performance in identifying benign lesions, its ability to detect malignant cases remains limited. These findings indicate that MobileNetV2 has potential as a clinical decision support system for skin cancer classification, but further improvements are required to enhance sensitivity toward malignant lesions.]]>
Copyrights © 2026
