<![CDATA[Melanoma is one of the most aggressive types of skin cancer with a high mortality rate if not detected at an early stage. In primary healthcare facilities, the lack of dermoscopy equipment causes examinations to rely solely on visual assessment, which may lead to diagnostic errors, particularly false negatives. This study aims to develop a web-based early melanoma detection system as a tool to assist initial screening. The proposed method implements a deep learning model based on the MobileNetV2 architecture using a transfer learning approach with pre-trained ImageNet weights. The dataset used in this study consists of melanoma and notmelanoma images from HAM10000, while the nonskin class is obtained from CIFAR-10 to help the model distinguish between skin lesion images and non-skin images. The dataset is divided into 70% training data, 20% validation data, and 10% testing data. Evaluation results show that the model achieves an accuracy of 90% in multiclass classification, while binary evaluation focusing on melanoma detection yields an accuracy of 90.48%, precision of 81.75%, recall of 91.96%, and an F1-score of 86.50% on the test data. The model is then implemented in a web-based system capable of displaying skin lesion classification results along with a confidence score in real time. The findings indicate that the developed system can perform automated image analysis and has the potential to be used as a supporting tool for early melanoma screening.]]>
